Research library
Agent Memory
Long-horizon memory for LLM agents: storage, consolidation, and forgetting.
158 documents
← All libraries-
Abstract
Long-running LLM agents accumulate interaction histories far larger than any context window, forcing a standing decision: what to encode deeply, what to forget, and what to retrieve under a fixed memory budget. Production systems answer with semantic similarity or recency -- both mis-specified for the forgetting decision, which is made at consolidation time before the future query is known. We propose a multi-factor memory value function V(m)=\sum_i w_i f_i(m) over seven interpretable factors (emotional intensity, goal relevance, value alignment, self/user relevance, task utility, reliability, and usage history) drawn from cognitive psychology, whose weights are learned from a downstream objective by a gradient-free optimiser, and whose single scalar uniformly controls encoding depth, forget risk, and retrieval rank. We make a methodological point: on LongMemEval, scoring goal relevance against the held-out evaluation question saturates gold-evidence retention at \approx 0.98 -- this measures retrieval, not forgetting. In the realistic blind regime, a learned multi-factor value retains 0.770 \pm 0.011 of gold evidence across 479 usable cases, versus 0.657 for uniform weights, 0.518 for the best single factor, and 0.368 for recency; every paired gap's 95% bootstrap CI is above zero, and a neural network over the same factors ties the linear model. The learned weights are interpretable -- reliability, emotional intensity, and self/user relevance dominate, while query-time goal similarity is correctly down-weighted for the forgetting decision. A controlled synthetic task with planted confounds confirms the learner recovers a separating weighting (1.00 retention) where uniform weighting fails (0.62). The substrate is open-source; all experiments run on a single CPU with no API calls.
-
Abstract
While Large Language Models (LLMs) have advanced open-domain dialogue systems, maintaining long-term consistency remains a challenge due to inherent limitations in long-context reasoning and the inefficiency of processing extensive raw text. Existing approaches typically rely on either unstructured memory storage, which is prone to information loss, or computationally expensive LLMs that incur high latency. To address these limitations, we propose G-Long, a graph-enhanced framework that utilizes a fine-tuned small Language Model (sLM) for structured triplet extraction and associative retrieval, significantly reducing operational costs. Furthermore, we introduce the novel attention-aware importance scoring mechanism that leverages the intrinsic cross-attention signals of a T5 summarizer to identify salient memories. Extensive experiments across diverse benchmarks demonstrate that G-Long achieves state-of-the-art performance in both response generation and memory retrieval, yielding performance gains of up to 9.8% in response quality on MSC and 40.8% in retrieval recall on LME, while significantly minimizing computational overhead.
-
Abstract
Long-term LLM agents need persistent memory that can track changing facts and provide relevant evidence across sessions. Existing memory systems often store observations as isolated records, summaries, or indexed fragments, which makes evidence aggregation, fact revision, and memory maintenance difficult. We propose Infini Memory, a maintainable text-based persistent memory architecture that treats agent memory as topic-structured documents. Each topic document serves as a semantic unit for collecting related evidence, preserving metadata, and revising facts over time. New observations are first staged in a buffer and periodically consolidated into coherent textual contexts. At inference time, an agentic retrieval procedure lets the LLM read memory through iterative tool calls rather than a single retrieval step. On MemoryAgentBench, Infini Memory achieves 64.7% overall score. Ablations show that topic-structured maintenance and iterative evidence inspection improve complementary aspects of long-term memory use.
-
Abstract
Long-term memory is the missing layer for LLM agents: across sessions they forget, and the common workaround -- replaying the whole history into the prompt -- is expensive, slow, and, as distractors accumulate, less accurate. Most memory systems win on cost or latency but still lose to the full-context baseline on accuracy, and benchmark numbers are reported on inconsistent, non-reproducible harnesses, so one system appears at wildly different scores across sources. We present Engram, an open-source, dual-process memory engine on a bi-temporal data model. A fast write path appends lossless episodes with no LLM on the critical path; an asynchronous path extracts atomic (subject, predicate, object) facts, builds a bi-temporal knowledge graph, and resolves contradictions without an LLM call per fact -- invalidating, never deleting, so every fact keeps provenance and a supersession chain. A hybrid read path fuses dense, lexical, graph, and recency/salience signals, applies a point-in-time ("as-of") filter, and assembles a compact, provenance-tagged context. On the full 500-question LongMemEval_S, graded by the official category-specific judge, Engram's lean configuration -- answering from a ~9.6k-token retrieved slice, never the full history -- scores 83.6% vs. 73.2% for full-context (+10.4 points, McNemar p < 10^-6) at ~8x fewer tokens (9.6k vs. 79k), with 0/500 errored. The gain needs a hybrid read path: facts alone lose recall, while facts plus retrieved chunks recover detail. We also contribute a neutral, in-repo evaluation harness with the official judge baked in and the full-context baseline in every table, publish the raw per-question logs, and document the measurement-integrity pitfalls (truncation, home-grown judges, full-history leaks) that silently distort memory benchmarks. Every number ships with a command to reproduce it.
-
Abstract
Long-horizon language agents accumulate observations, reasoning traces, and retrieved facts that exceed their finite context windows, making memory retention a fundamental resource-allocation problem. Existing memory systems improve management through heuristic scoring, retrieval optimization, or learned compression, but largely treat retention as a local decision problem and do not explicitly model its long-term consequences under realistic observability constraints. To fill this gap, we formulate memory retention as a constrained stochastic optimization problem with explicit budget feasibility, evidence utility, and delayed costs including miss penalties, reacquisition delays, and stale-information risk. We then propose OSL-MR (Observability-Safe Learning for Memory Retention), a novel framework that enforces a strict separation between online-observable features and offline-available supervision (OAS). OSL-MR combines an evidence learner trained from realized evidence supervision with a Mixed-Score heuristic that serves both as a deployable online-safe baseline and as a structured inductive prior for learning. The resulting policy learns query-conditioned evidence value directly from interaction data while remaining deployable under the same observability constraints. Experiments on LOCOMO and LongMemEval show that OSL-MR consistently outperforms recency-based methods, Generative Agents-style scoring, and other heuristic baselines, particularly under tight memory budgets. The Mixed-Score prior further improves precision while preserving recall, and sensitivity analysis demonstrates robustness across a wide range of cost configurations.
-
Abstract
Memory is essential for enabling large language models to support long-horizon reasoning, yet existing memory systems remain unreliable and difficult to debug. Tracing memory's dynamic evolution is crucial to understand how information is synthesized, propagated, or corrupted over time. In this work, we study the new problem of error tracing and attribution in LLM memory systems. We propose a novel framework that transforms memory pipelines into executable memory evolution graphs, enabling fine-grained tracing of operational information flow. We then construct MemTraceBench, a benchmark collected from representative memory systems such as Long-Context, RAG, Mem0, and EverMemOS, to systematically study memory failure modes. We further introduce an automatic attribution method that iteratively traces operation subgraphs to pinpoint the root cause of any failed case. Our analysis reveals that memory failures are systematic, stemming from operation-level issues like information loss and retrieval misalignment. Crucially, we leverage these fine-grained attribution signals to guide downstream prompt optimization, establishing a closed-loop system that automatically corrects faults and boosts end-task performance by up to 7.62%. Code will be released at https://github.com/zjunlp/MemTrace.
-
Abstract
Core systems like key-value stores have historically taken years to build, and are designed to be general so as to amortize cost across deployments, paying a significant performance cost. We argue that LLM-based coding agents now make a different approach tractable: Just-in-Time Systems, in which the entire system is synthesized from scratch, specialized to the environment, workload, and required system properties. We present a JIT system synthesis pipeline, Jitskit, and explore its effectiveness in synthesizing key-value stores from spec cards that span different YCSB workloads, deployment constraints (e.g., compute resources), and system properties (e.g., consistency and durability). Jitskit iteratively refines a system implementation to match the specification against an evolving evaluation test suite. The resulting synthesized systems are performant, beating comparable state-of-the-art systems on 18 of 18 specs tried, by up to 4.6x over the best off-the-shelf baseline on the most favorable spec. Naively running Claude Code either reward-hacks or underperforms Jitskit by up to 5.4x. We discuss the challenges we overcame in building Jitskit and our key takeaways.
-
Abstract
To enable reliable long-term interaction, LLM agents require a memory system that can faithfully store, efficiently retrieve, and deeply reason over accumulated dialogue history. Most existing methods adopt an extracted fact based paradigm: handcrafted static prompts compress raw dialogues into atomic facts, which are then stored, matched, and injected into downstream reasoning. Nevertheless, such fact-centric designs inevitably discard fine-grained details in original dialogues and fail to support deep reasoning over scattered isolated facts. Moreover, static prompts cannot maintain consistent extraction granularity across diverse dialogue styles. To address these limitations, we propose TriMem, which maintains three coexisting representation granularities, including raw dialogue segments anchored by source identifiers for storage fidelity, extracted atomic facts for efficient memory retrieval, synthesized profiles that aggregate dispersed facts into holistic semantic understanding for deep reasoning. We further adopt TextGrad-based prompt optimization, which iteratively refines extraction and profiling prompts via response quality feedback, achieving lifelong evolution without any parameter updating. Extensive experiments on LoCoMo and PerLTQA across multiple LLM backbones demonstrate that TriMem consistently outperforms strong memory baselines. The code is available at https://TMLR-TriMem.github.io .
-
Abstract
LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear whether a model's base capability in task-solving predicts its capabilities in harness self-evolution: which models produce useful harness updates, and which actually benefit from them? We analyze two harness self-evolution capabilities: (i) harness-updating, the capability to produce useful persistent harness updates from execution evidence; (ii) harness-benefit, the capability to benefit from updated harnesses during task solving. Our analysis reveals two findings. First, harness-updating is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gains; even Qwen3.5-9B's updates yield gains comparable to those of Claude Opus~4.6. Second, harness-benefit is non-monotonic in base capability: weak-tier models benefit little from updated harnesses, mid-tier models benefit most, and strong-tier models benefit less than mid-tier. We trace low gains at the weak tier to two failure modes: weak-tier models may fail to activate relevant harness artifacts, or activate them but fail to follow them faithfully. These findings suggest investing capability budget in the task-solving agent rather than the evolver, and targeting harness invocation and long-horizon instruction following in agent training. Our source code is publicly available at https://github.com/A-EVO-Lab/a-evolve/tree/release/harness-evolution.
-
Abstract
Long-lived AI agents are increasingly deployed as persistent operational systems, yet they are still evaluated like freshly initialized models. Day-one benchmarks miss a basic systems question: how long does an agent remain reliable after deployment? Even when model weights are frozen, an agent's effective state keeps changing as it compresses interaction history, retrieves from a growing memory store, revises facts after updates, and undergoes routine maintenance. Reliability therefore becomes a lifespan property of the full agent harness, not only a snapshot property of the base model. We introduce AgingBench, a longitudinal reliability benchmark for agent lifespan engineering: measuring not only whether deployed agents degrade, but what form the degradation takes and where repair should target. AgingBench organizes agent aging into four mechanisms: compression aging, interference aging, revision aging, and maintenance aging. To diagnose these failures, AgingBench uses temporal dependency graphs and paired counterfactual probes that produce diagnostic profiles for the write, retrieval, and utilization stages of the memory pipeline. Across 7 scenarios, 14 models, multiple memory policies, and both runner-controlled and autonomous agents, over ~400 runs spanning 8 - 200 sessions show that agent aging is not one-dimensional: behavioral tests can remain clean while factual precision decays; derived-state tracking can collapse sharply within a single model; and the same wrong answer can require different repairs depending on what the diagnostic profile points to. These results suggest that reliable agent deployment requires lifespan evaluation, mechanism-level diagnosis, and stage-targeted repair, not only stronger day-one models.
-
Abstract
Long-horizon conversational agents rely on memory systems with increasingly sophisticated retrieval mechanisms. However, retrieved fragments are typically fed to the language model as unstructured text, lacking the relational, temporal, and thematic structures essential for complex reasoning. To bridge this reasoning gap, we introduce GRAVITY (\textbf{G}eneration-time \textbf{R}elational \textbf{A}nchoring \textbf{V}ia \textbf{I}njected \textbf{T}opological Memor\textbf{Y}), a plug-and-play structured memory module. GRAVITY extracts three complementary knowledge representations from raw conversational utterances: entity profiles grounded in relational graphs, temporal event tuples linked into causal traces, and cross-session topic summaries. At generation time, it injects these representations into the host system's prompt as structured anchoring contexts. This approach effectively synthesizes scattered evidence into a coherent, query-relevant context without requiring any architectural modifications to the host model. Extensive evaluations across five diverse memory systems on the LongMemEval and LoCoMo benchmarks demonstrate the efficacy of our approach. On average, GRAVITY improves LLM-judge accuracy by 7.5--10.1%. Gains are inversely correlated with baseline strength: the weakest host improves by 12.2% while the strongest still gains 3.8--5.7%. These findings establish structured context anchoring as a broadly effective, architecture-agnostic augmentation paradigm for long-horizon conversational memory.
-
Abstract
Large Language Model (LLM)-based agents have fundamentally reshaped artificial intelligence by integrating external tools and planning capabilities. While memory mechanisms have emerged as the architectural cornerstone of these systems, current research remains fragmented, oscillating between operating system engineering and cognitive science. This theoretical divide prevents a unified view of technological synthesis and a coherent evolutionary perspective. To bridge this gap, this survey proposes a novel evolutionary framework for LLM agent memory mechanisms, formalizing the development process into three stages: Storage (trajectory preservation), Reflection (trajectory refinement), and Experience (trajectory abstraction). We first formally define these three stages before analyzing the three core drivers of this evolution: the necessity for long-range consistency, the challenges in dynamic environments, and the ultimate goal of continual learning. Furthermore, we specifically explore two transformative mechanisms in the frontier Experience stage: proactive exploration and cross-trajectory abstraction. By synthesizing these disparate views, this work offers robust design principles and a clear roadmap for the development of next-generation LLM agents.
-
Abstract
Large Language Model (LLM) agents are increasingly expected to maintain coherent, long-term personalized memory, yet current benchmarks primarily measure static fact retrieval, overlooking the ability to revise stored beliefs when new evidence emerges. We identify a critical and underexplored failure mode, Implicit Conflict: a later observation invalidates an earlier memory without explicit negation, requiring contextual inference and commonsense reasoning to detect. To rigorously evaluate this capability, we introduce STALE, a benchmark of 400 expert-validated conflict scenarios (1,200 evaluation queries across three probing dimensions) spanning over 100 everyday topics with contexts up to 150K tokens. We propose a three-dimensional probing framework that tests State Resolution (detecting that a prior belief is outdated), Premise Resistance (rejecting queries that falsely presuppose a stale state), and Implicit Policy Adaptation (proactively applying updated states in downstream behavior). A systematic evaluation of frontier LLMs and specialized memory frameworks reveals a pervasive gap between retrieving updated evidence and acting on it, with even the best evaluated model achieving only 55.2% overall accuracy. Models often accept outdated assumptions embedded in a user's query, and they struggle to recognize when a change in one aspect of the user's state should invalidate related memories. To establish an initial baseline for state-aware memory, we further present CUPMem, a prototype that strengthens write-time revision through structured state consolidation and propagation-aware search, suggesting that explicit state adjudication is a promising direction for robust agentic memory.
-
Abstract
Large language model (LLM) agents increasingly leverage long term memory to support persistent and autonomous task execution. However, this capability also introduces a new attack surface: memory poisoning, where adversaries can inject malicious information to influence future behavior. Existing memory poisoning attacks often assume that injected content can be stored directly in memory, overlooking the selective extraction and rewriting stages in modern memory pipelines. This makes prior methods ineffective under realistic settings. In this paper, we propose MemPoison, a novel memory poisoning attack that bypasses selective memory mechanisms in LLM agents, where an attacker can inject triggerable backdoors into the agent's long-term memory through dialogue interactions, thereby misleading its subsequent responses. MemPoison introduces three key components: (i) a semantic relational bridge that binds the trigger and payload into a coherent statement to ensure they are extracted into memory together; (ii) entity masquerading that optimizes triggers to mimic named entities, resisting rewriting; and (iii) joint embedding optimization that shapes trigger-injected texts into a tight cluster in the embedding space while maintaining isolation from benign embeddings for stealth. Evaluations across different agent domains and memory mechanisms show MemPoison achieves attack success rates up to 0.95, outperforming existing baselines. Mechanistic analysis indicates that the attack exploits embedding-space anisotropy and shifts attention patterns, highlighting core vulnerabilities in selective memory systems. We evaluate multiple defense strategies and demonstrate their fundamental limitations in mitigating the attack.
-
Abstract
Long-term memory systems enable conversational agents based on large language models (LLMs) to retain, retrieve, and apply user-specific information across multi-session interactions. However, existing evaluations mainly assess outcome-level performance or temporal updating, providing limited insight into how systems retrieve and rank temporally valid, factually correct, and contextually applicable memory evidence under conflicting alternatives. To address this gap, we propose MemConflict, a diagnostic framework that treats memory validity as a query-conditioned fitness-for-use problem. MemConflict formalizes dynamic, static, and conditional conflicts over temporal validity, factual correctness, and contextual applicability. It simulates controlled long-horizon histories from structured user profiles, introduces cross-session conflicts, and injects semantically similar distractors to create competition among memory candidates. The resulting multi-session dialogue benchmark supports black-box evaluation of final answers and white-box analysis of supporting-memory retrieval and ranking. Experiments on six representative long-term memory systems show uneven strengths across conflict types, with answer correctness often diverging from memory retrieval and ranking. Sensitivity analyses reveal that longer histories, distractors, implicit queries, and larger conflict distances degrade performance. Diagnostics show failures from missing supporting memories and ineffective use of retrieved memories. Collectively, MemConflict advances principled long-term memory governance through retrieval-aware, conflict-aware reliability assessment.
-
Abstract
Conversational-memory systems increasingly transform dialogue history into facts, summaries, timelines, and other source-linked descendants, so a single source turn can coexist with several derived memories in the same retrieval index. This raises an underspecified evaluation question: which stored form should receive retrieval credit? We show that this scoring-target choice is often left implicit and can materially change benchmark conclusions. We present TIAP, a fixed-output audit that rescores saved ranked outputs under three targets -- Raw, Source, and Canonical -- without rerunning retrieval. On LoCoMo and LongMemEval-S, switching only the credited target changes nDCG on 83.4--94.0 percent of shared queries, flips target orderings on Mem0 and MemoryOS transfer runs, and reverses parser-density recommendations. A 1,902-case semantic audit further shows that relaxed source-linked credit is fully justified only 29.2 percent of the time, despite high rubric reliability in a validation subset. These results reveal target noninvariance: conclusions about memory architectures can silently flip with a single benchmark-design choice. Conversational-memory papers should therefore define and report the scoring target explicitly.
-
Abstract
Large language model (LLM) agents accumulate rich episodic trajectories while solving real-world tasks, but it remains unclear whether such experience can be distilled into reusable procedural skills. We introduce SkillEvolBench, a diagnostic benchmark for evaluating this step from experience reuse to skill formation. It contains 180 tasks across six real-world agent environments, organized into role-conditioned task families with shared latent procedures. Agents learn from acquisition tasks, update an external skill library using compacted trajectories and verifier feedback, and then face frozen deployment tasks testing context shift, adversarial shortcuts, and composition. By comparing self-generated and curated-start skill evolution against no-skill and raw-trajectory controls, SkillEvolBench separates procedural abstraction from base capability, curated prior knowledge, and direct reuse of episodic traces. Across ten model configurations and three agent harnesses, we find that current agents often adapt locally but rarely form robust reusable skills. Skill-based conditions can improve acquisition or replay, and individual models sometimes gain on specific deployment axes, but these gains are unstable under frozen deployment. Raw-trajectory reuse frequently outperforms distilled skills, suggesting that current abstraction procedures discard contextual and procedural cues that remain useful for future tasks. Capacity and cost analyses further show that writing more skills or larger Tier-3 resource libraries is not sufficient: additional updates can improve coverage while introducing episode-specific drift and procedural clutter. These findings position SkillEvolBench as a testbed for measuring when one-off experience becomes durable procedural knowledge rather than task-local memory.
-
Abstract
Large Language Models (LLMs) struggle with knowledge-intensive tasks due to hallucinations and fragmented reasoning over dispersed information. While Retrieval-Augmented Generation (RAG) grounds generation in external sources, existing methods often treat evidence as isolated units, failing to reconstruct the logical chains that connect these dots. Inspired by Complementary Learning Systems (CLS), we propose CodaRAG, a framework that evolves retrieval from passive lookup into active associative discovery. CodaRAG operates via a three-stage pipeline: (1) Knowledge Consolidation to unify fragmented extractions into a stable memory substrate; (2) Associative Navigation to traverse the graph via multi-dimensional pathways-semantic, contextualized, and functional-explicitly recovering dispersed evidence chains; and (3) Interference Elimination to prune hyper-associative noise, ensuring a coherent, high-precision reasoning context. On GraphRAG-Bench, CodaRAG achieves absolute gains of 7-10% in retrieval recall and 3-11% in generation accuracy. These results demonstrate CodaRAG's superior ability to systematically robustify associative evidence retrieval for factual, reasoning, and creative tasks.
-
Abstract
The two most influential cognitive architecture frameworks for AI agents, CoALA [21] and JEPA [12], both lack an explicit Knowledge layer with its own persistence semantics. This gap produces a category error: systems apply cognitive decay to factual claims, or treat facts and experiences with identical update mechanics. We survey persistence semantics across existing memory systems and identify eight convergence points, from Karpathy's LLM Knowledge Base [10] to the BEAM benchmark's near-zero contradiction-resolution scores [22], all pointing to related architectural gaps. We propose a four-layer decom position (Knowledge, Memory, Wisdom, Intelligence) where each layer has fundamentally different persistence semantics: indefinite supersession, Ebbinghaus decay, evidence-gated revision, and ephemeral inference respectively. Companion implementations in Python and Rust demonstrate the architectural separation is feasible. We borrow terminology from cognitive science as a useful analogy (the Knowledge/Memory distinction echoes Tulving's trichotomy), but our layers are engineering constructs justified by persistence-semantics requirements, not by neural architecture. We argue that these distinctions demand distinct persistence semantics in engineering implementations, and that no current framework or system provides this.
-
Abstract
Embodied agents are expected to operate persistently in dynamic physical environments, continuously acquiring new capabilities over time. Existing approaches to improving agent performance often rely on modifying the agent itself -- through prompt engineering, policy updates, or structural redesign -- leading to instability and loss of identity in long-lived systems. In this work, we propose a capability-centric evolution paradigm for embodied agents. We argue that a robot should maintain a persistent agent as its cognitive identity, while enabling continuous improvement through the evolution of its capabilities. Specifically, we introduce the concept of Embodied Capability Modules (ECMs), which represent modular, versioned units of embodied functionality that can be learned, refined, and composed over time. We present a unified framework in which capability evolution is decoupled from agent identity. Capabilities evolve through a closed-loop process involving task execution, experience collection, model refinement, and module updating, while all executions are governed by a runtime layer that enforces safety and policy constraints. We demonstrate through simulated embodied tasks that capability evolution improves task success rates from 32.4% to 91.3% over 20 iterations, outperforming both agent-modification baselines and established skill-learning methods (SPiRL, SkiMo), while preserving zero policy drift and zero safety violations. Our results suggest that separating agent identity from capability evolution provides a scalable and safe foundation for long-term embodied intelligence.
-
Abstract
Long-term memory is essential for conversational agents to maintain coherence, track persistent tasks, and provide personalized interactions across extended dialogues. However, existing approaches as Retrieval-Augmented Generation (RAG) and graph-based memory mostly rely on pairwise relations, which can hardly capture high-order associations, i.e., joint dependencies among multiple elements, causing fragmented retrieval. To this end, we propose HyperMem, a hypergraph-based hierarchical memory architecture that explicitly models such associations using hyperedges. Particularly, HyperMem structures memory into three levels: topics, episodes, and facts, and groups related episodes and their facts via hyperedges, unifying scattered content into coherent units. Leveraging this structure, we design a hybrid lexical-semantic index and a coarse-to-fine retrieval strategy, supporting accurate and efficient retrieval of high-order associations. Experiments on the LoCoMo benchmark show that HyperMem achieves state-of-the-art performance with 92.73% LLM-as-a-judge accuracy, demonstrating the effectiveness of HyperMem for long-term conversations.
-
Abstract
Structured memory representations such as knowledge graphs are central to autonomous agents and other long-lived systems. However, most existing approaches model time as discrete metadata, either sorting by recency (burying old-yet-permanent knowledge), simply overwriting outdated facts, or requiring an expensive LLM call at every ingestion step, leaving them unable to distinguish persistent facts from evolving ones. To address this, we introduce RoMem, a drop-in temporal knowledge graph module for structured memory systems, applicable to agentic memory and beyond. A pretrained Semantic Speed Gate maps each relation's text embedding to a volatility score, learning from data that evolving relations (e.g., "president of") should rotate fast while persistent ones (e.g., "born in") should remain stable. Combined with continuous phase rotation, this enables geometric shadowing: obsolete facts are rotated out of phase in complex vector space, so temporally correct facts naturally outrank contradictions without deletion. On temporal knowledge graph completion, RoMem achieves state-of-the-art results on ICEWS05-15 (72.6 MRR). Applied to agentic memory, it delivers 2-3x MRR and answer accuracy on temporal reasoning (MultiTQ), dominates hybrid benchmark (LoCoMo), preserves static memory with zero degradation (DMR-MSC), and generalises zero-shot to unseen financial domains (FinTMMBench).
-
Abstract
While large language model--powered agents can self-evolve by accumulating experience or by dynamically creating new assets (i.e., tools or expert agents), existing frameworks typically treat these two evolutionary processes in isolation. This separation overlooks their intrinsic interdependence: the former is inherently bounded by a manually predefined static toolset, while the latter generates new assets from scratch without experiential guidance, leading to limited capability growth and unstable evolution. To address this limitation, we introduce a novel paradigm of co-evolutionary Capability Expansion and Experience Distillation. Guided by this paradigm, we propose the \textbf{Mem$^{\textbf{2}}$Evolve}, which integrates two core components: \textbf{Experience Memory} and \textbf{Asset Memory}. Specifically, Mem$^{2}$Evolve leverages accumulated experience to guide the dynamic creation of assets, thereby expanding the agent's capability space while simultaneously acquiring new experience to achieve co-evolution. Extensive experiments across 6 task categories and 8 benchmarks demonstrate that Mem$^{2}$Evolve achieves improvement of 18.53\% over standard LLMs, 11.80\% over agents evolving solely through experience, and 6.46\% over those evolving solely through asset creation, establishing it as a substantially more effective and stable self-evolving agent framework. Code is available at: https://buaa-irip-llm.github.io/Mem2Evolve.
-
Abstract
Personalized large language models (PLLMs) have garnered significant attention for their ability to align outputs with individual's needs and preferences. However, they still struggle with long-horizon tasks, such as tracking a user's extensive history of conversations or activities. Existing memory mechanisms often fail to capture evolving behaviors, and RAG paradigms are trapped by a quality-efficiency tradeoff. Meanwhile, parametric adaptation is bottlenecked by train-inference gap due to the scarcity of labeled data. To enhance the long-horizon capabilities of PLLMs, we introduce TSUBASA, a two-pronged approach designed to improve memory writing via dynamic memory evolution, and memory reading via self-learning with a context distillation objective to internalize user experiences. Extensive evaluations on long-horizon benchmarks using the Qwen-3 model family (4B to 32B) validate the effectiveness of TSUBASA, surpassing competitive memory-augmented systems that rely primarily on memory writing, such as Mem0 and Memory-R1. Our analyses further confirms that TSUBASA breaks the quality-efficiency barrier to achieve Pareto improvements, delivering robust, high-fidelity personalization with a reduced token budget.
-
Abstract
Autonomous AI agents operating in dynamic environments face a persistent challenge: acquiring new capabilities without erasing prior knowledge. We present Adaptive Memory Crystallization (AMC), a memory architecture for progressive experience consolidation in continual reinforcement learning. AMC is conceptually inspired by the qualitative structure of synaptic tagging and capture (STC) theory, the idea that memories transition through discrete stability phases, but makes no claim to model the underlying molecular or synaptic mechanisms. AMC models memory as a continuous crystallization process in which experiences migrate from plastic to stable states according to a multi-objective utility signal. The framework introduces a three-phase memory hierarchy (Liquid--Glass--Crystal) governed by an Itô stochastic differential equation (SDE) whose population-level behavior is captured by an explicit Fokker--Planck equation admitting a closed-form Beta stationary distribution. We provide proofs of: (i) well-posedness and global convergence of the crystallization SDE to a unique Beta stationary distribution; (ii) exponential convergence of individual crystallization states to their fixed points, with explicit rates and variance bounds; and (iii) end-to-end Q-learning error bounds and matching memory-capacity lower bounds that link SDE parameters directly to agent performance. Empirical evaluation on Meta-World MT50, Atari 20-game sequential learning, and MuJoCo continual locomotion consistently shows improvements in forward transfer (+34--43\% over the strongest baseline), reductions in catastrophic forgetting (67--80\%), and a 62\% decrease in memory footprint.
-
Abstract
Large language model assistants are increasingly expected to retain and reason over information accumulated across many sessions. We introduce EngramaBench, a benchmark for long-term conversational memory built around five personas, one hundred multi-session conversations, and one hundred fifty queries spanning factual recall, cross-space integration, temporal reasoning, adversarial abstention, and emergent synthesis. We evaluate Engrama, a graph-structured memory system, against GPT-4o full-context prompting and Mem0, an open-source vector-retrieval memory system. All three use the same answering model (GPT-4o), isolating the effect of memory architecture. GPT-4o full-context achieves the highest composite score (0.6186), while Engrama scores 0.5367 globally but is the only system to score higher than full-context prompting on cross-space reasoning (0.6532 vs. 0.6291, n=30). Mem0 is cheapest but substantially weaker (0.4809). Ablations reveal that the components driving Engrama's cross-space advantage trade off against global composite score, exposing a systems-level tension between structured memory specialization and aggregate optimization.
-
Abstract
Current LLM-based agents demonstrate strong performance in episodic task execution but remain constrained by static toolsets and episodic amnesia, failing to accumulate experience across task boundaries. This paper formalizes the Self-Evolving Agent (SEA) from the perspective of digital embodiment and continuous cross-task evolution, introduces the Evolutionary Flywheel as its minimal sufficient architecture, and presents SEA-Eval -- the first benchmark designed specifically for evaluating SEAs. Grounded in Flywheel theory, SEA-Eval establishes SR and T as primary metrics and, through sequential task stream design, is designed to quantify evolutionary gain, evolutionary stability, and implicit alignment convergence. Empirical evaluation reveals that, under comparable success rates, token consumption differs by up to 31.2 times between frameworks on individual tasks, with divergent evolutionary trajectories emerging under sequential analysis -- demonstrating that success rate alone creates a capability illusion and that the sequential convergence of $T$ is the key criterion for distinguishing genuine evolution from pseudo-evolution.
-
Abstract
Robots must verbalize their past experiences when users ask "Where did you put my keys?" or "Why did the task fail?" Yet maintaining life-long episodic memory (EM) from continuous multimodal perception quickly exceeds storage limits and makes real-time query impractical, calling for selective forgetting that adapts to users' notions of relevance. We present H$^2$-EMV, a framework enabling humanoids to learn what to remember through user interaction. Our approach incrementally constructs hierarchical EM, selectively forgets using language-model-based relevance estimation conditioned on learned natural-language rules, and updates these rules given user feedback about forgotten details. Evaluations on simulated household tasks and 20.5-hour-long real-world recordings from ARMAR-7 demonstrate that H$^2$-EMV maintains question-answering accuracy while reducing memory size by 45% and query-time compute by 35%. Critically, performance improves over time - accuracy increases 70% in second-round queries by adapting to user-specific priorities - demonstrating that learned forgetting enables scalable, personalized EM for long-term human-robot collaboration.
-
Abstract
Regulated enterprises in the United States -- banks, telecommunications providers, large technology companies -- operate across heterogeneous systems that were rarely designed to interoperate. ERP platforms, billing engines, supply chain tools, and financial reporting infrastructure coexist within the same organization, but they do not talk to each other well. The resulting fragmentation produces familiar problems: transactions recorded in one system but unreconciled in another, asset inventories drifting from their systems of record, and audit-readiness that depends on manual effort. The PCAOB's 2024 inspection cycle put a number on the consequences: a 39% aggregate Part I.A deficiency rate across all inspected firms. This paper introduces the GERA Framework (Governed Enterprise Reconciliation Architecture) -- a vendor-neutral, four-layer data architecture that integrates deterministic cross-system reconciliation, statistical anomaly detection (baseline Z-Score with robust alternatives), governed semantic standardization, and NIST CSF 2.0-aligned security controls into a single methodology. The architecture spans four layers (ingestion, staging, core models, and semantic serving), following the multi-layer pattern now common in modern data platforms. The patterns are demonstrated through U.S. broadband operations -- where billing reconciliation, inventory aging, and governance are tightly coupled -- and draw on the author's implementation experience across three regulated enterprise environments: a regional bank, a national broadband provider, and a Fortune 500 technology company's central finance organization. This is a practitioner reference -- an architectural framework paper documenting field-tested patterns -- not a controlled experiment or benchmark study. No proprietary systems, datasets, or internal implementations are disclosed.
-
Abstract
Agentic AI systems - systems that can pursue goals through multi-step planning and tool-mediated action with limited direct supervision - are moving from experimental prototypes to enterprise deployments. This transition introduces tensions in implementation, scaling, and governance: organizations seek scalable autonomy for knowledge and coordination work, yet must preserve accountability, safety, cost control, and responsibility as systems initiate actions, access enterprise data, and evolve through iterative updates. Building on an in-depth qualitative case of a large IT services company's 2025 development and staged rollout of an agentic system integrated with enterprise tools; we show that governance is implemented through concrete architectural and working arrangements that determine what the system is allowed to do, which tools and data it can use, how memory is handled, and how performance improvements are introduced over time. We then distill seven lessons that explain how to build effective governance into agentic AI during operationalization and scaling.
-
Abstract
Large language model (LLM) agents such as OpenClaw rely on reusable skills to perform complex tasks, yet these skills remain largely static after deployment. As a result, similar workflows, tool usage patterns, and failure modes are repeatedly rediscovered across users, preventing the system from improving with experience. While interactions from different users provide complementary signals about when a skill works or fails, existing systems lack a mechanism to convert such heterogeneous experiences into reliable skill updates. To address these issues, we present SkillClaw, a framework for collective skill evolution in multi-user agent ecosystems, which treats cross-user and over-time interactions as the primary signal for improving skills. SkillClaw continuously aggregates trajectories generated during use and processes them with an autonomous evolver, which identifies recurring behavioral patterns and translates them into updates to the skill set by refining existing skills or extending them with new capabilities. The resulting skills are maintained in a shared repository and synchronized across users, allowing improvements discovered in one context to propagate system-wide while requiring no additional effort from users. By integrating multi-user experience into ongoing skill updates, SkillClaw enables cross-user knowledge transfer and cumulative capability improvement, and experiments on WildClawBench show that limited interaction and feedback, it significantly improves the performance of Qwen3-Max in real-world agent scenarios.
-
Abstract
Large language model agents rely on effective model context to obtain task-relevant information for decision-making. Many existing context engineering approaches primarily rely on the context generated from the past experience and retrieval mechanisms that reuse these context. However, retrieved context from past tasks must be adapted by the execution agent to fit new situations, placing additional reasoning burden on the underlying LLM. To address this limitation, we propose a generative context augmentation framework using Contrastive Learning of Experience via Agentic Reflection (CLEAR). CLEAR first employs a reflection agent to perform contrastive analysis over past execution trajectories and summarize useful context for each observed task. These summaries are then used as supervised fine-tuning data to train a context augmentation model (CAM). Then we further optimize CAM using reinforcement learning, where the reward signal is obtained by running the task execution agent. By learning to generate task-specific knowledge rather than retrieve knowledge from the past, CAM produces context that is better tailored to the current task. We conduct comprehensive evaluations on the AppWorld and WebShop benchmarks. Experimental results show that CLEAR consistently outperforms strong baselines. It improves task completion rate from 72.62% to 81.15% on AppWorld test set and averaged reward from 0.68 to 0.74 on a subset of WebShop, compared with baseline agent. Our code is publicly available at https://github.com/awslabs/CLEAR.
-
Abstract
The emergence of Large Language Models (LLMs) has illuminated the potential for a general-purpose user simulator. However, existing benchmarks remain constrained to isolated scenarios, narrow action spaces, or synthetic data, failing to capture the holistic nature of authentic human behavior. To bridge this gap, we introduce OmniBehavior, the first user simulation benchmark constructed entirely from real-world data, integrating long-horizon, cross-scenario, and heterogeneous behavioral patterns into a unified framework. Based on this benchmark, we first provide empirical evidence that previous datasets with isolated scenarios suffer from tunnel vision, whereas real-world decision-making relies on long-term, cross-scenario causal chains. Extensive evaluations of state-of-the-art LLMs reveal that current models struggle to accurately simulate these complex behaviors, with performance plateauing even as context windows expand. Crucially, a systematic comparison between simulated and authentic behaviors uncovers a fundamental structural bias: LLMs tend to converge toward a positive average person, exhibiting hyper-activity, persona homogenization, and a utopian bias. This results in the loss of individual differences and long-tail behaviors, highlighting critical directions for future high-fidelity simulation research.
-
Abstract
Personal AI tools can now be generated from natural-language requests, but they often remain isolated after creation. We present PSI, a shared-state architecture that turns independently generated modules into coherent instruments: persistent, connected, and chat-complementary artifacts accessible through both GUIs and a generic chat agent. By publishing current state and write-back affordances to a shared personal-context bus, modules enable cross-module reasoning and synchronized actions across interfaces. We study PSI through a three-week autobiographical deployment in a self-developed personal AI environment and show that later-generated instruments can be integrated automatically through the same contract. PSI identifies shared state as the missing systems layer that transforms AI-generated personal software from isolated apps into coherent personal computing environments.
-
Abstract
When LLM conversations grow beyond the context window, old content must be evicted -- but how does the model recover it when needed? We propose cooperative paging: evicted segments are replaced with minimal keyword bookmarks ([pN:keywords], ~8-24 tokens each), and the model is given a recall() tool to retrieve full content on demand. On the LoCoMo benchmark (10 real multi-session conversations, 300+ turns), cooperative paging achieves the highest answer quality among six methods -- outperforming truncation, BM25, word-overlap retrieval, a search-tool baseline, and full context -- on four models (GPT-4o-mini, DeepSeek-v3.2, Claude Haiku, GLM-5), confirmed by four independent LLM judges ($p=0.017$, paired bootstrap). We then study the paging design space with a 5x4 ablation over boundary strategies and eviction policies (3,176 synthetic probes, 1,600 LoCoMo probes). Key findings: (1) coarse fixed-size pages (fixed_20) reach 96.7% while content-aware topic_shift collapses to 56.7%; (2) eviction policy choice is data-dependent (FIFO best on synthetic, LFU on LoCoMo); (3) two bookmark generation strategies improve over the heuristic baseline (+4.4 and +8.7 E2E points); (4) the remaining bottleneck is bookmark discrimination -- the model triggers recall() 96% of the time but selects the correct page only 57% when bookmarks are insufficiently distinctive. Keyword specificity alone accounts for a 25 percentage point accuracy difference.
-
Abstract
Long-term memory is fundamental for personalized and autonomous agents, yet populating it remains a bottleneck. Existing systems treat memory extraction as a one-shot, passive transcription from context to structured entries, which struggles with noisy dialogue, missing references, and cross-turn dependencies, leading to memory pollution, low-value writes, and inconsistency. In this paper, we introduce the MemReader family for active long-term memory extraction in agent systems: MemReader-0.6B, a compact and cost-efficient passive extractor distilled for accurate and schema-consistent structured outputs, and MemReader-4B, an active extractor optimized with Group Relative Policy Optimization (GRPO) to make memory writing decisions. Under a ReAct-style paradigm, MemReader-4B explicitly evaluates information value, reference ambiguity, and completeness before acting, and can selectively write memories, defer incomplete inputs, retrieve historical context, or discard irrelevant chatter. Experiments on LOCOMO, LongMemEval, and HaluMem show that MemReader consistently outperforms existing extraction-based baselines. In particular, MemReader-4B achieves state-of-the-art performance on tasks involving knowledge updating, temporal reasoning, and hallucination reduction. These results suggest that effective agent memory requires not merely extracting more information, but performing reasoning-driven and selective memory extraction to build low-noise and dynamically evolving long-term memory. Furthermore, MemReader has been integrated into MemOS and is being deployed in real-world applications. To support future research and adoption, we release the models and provide public API access.
-
Abstract
Large language models (LLMs) have transformed AI research thanks to their powerful internal capabilities and knowledge. However, existing LLMs still fail to effectively incorporate the massive external knowledge when interacting with the world. Although retrieval-augmented LLMs are proposed to mitigate the issue, they are still fundamentally constrained by the context length of LLMs, as they can only retrieve top-K raw data chunks from the external knowledge base which often consists of millions of data chunks. Here we propose Thought-Retriever, a novel model-agnostic algorithm that helps LLMs generate output conditioned on arbitrarily long external data, without being constrained by the context length or number of retrieved data chunks. Our key insight is to let an LLM fully leverage its intermediate responses generated when solving past user queries (thoughts), filtering meaningless and redundant thoughts, organizing them in thought memory, and retrieving the relevant thoughts when addressing new queries. This effectively equips LLM-based agents with a self-evolving long-term memory that grows more capable through continuous interaction. Besides algorithmic innovation, we further meticulously prepare a novel benchmark, AcademicEval, which requires an LLM to faithfully leverage ultra-long context to answer queries based on real-world academic papers. Extensive experiments on AcademicEval and two other public datasets validate that Thought-Retriever remarkably outperforms state-of-the-art baselines, achieving an average increase of at least 7.6% in F1 score and 16% in win rate across various tasks. More importantly, we further demonstrate two exciting findings: (1) Thought-Retriever can indeed help LLM self-evolve after solving more user queries; (2) Thought-Retriever learns to leverage deeper thoughts to answer more abstract user queries.
-
Abstract
Large Language Model (LLM) agents have demonstrated remarkable capabilities in reasoning and tool use, yet they often suffer from rigid, reactive control flows that limit their adaptability and efficiency. Most existing frameworks rely on fixed pipelines or failure-triggered reflection, causing agents to act impulsively or correct errors only after they occur. In this paper, we introduce Heartbeat-Driven Autonomous Thinking Activity Scheduling, a mechanism that enables proactive, adaptive, and continuous self-regulation. Mirroring the natural rhythm of human cognition, our system employs a periodic ``heartbeat'' mechanism to orchestrate a dynamic repertoire of cognitive modules (e.g., Planner, Critic, Recaller, Dreamer). Unlike traditional approaches that rely on hard-coded symbolic rules or immediate reactive triggers, our scheduler learns to determine when to engage specific thinking activities -- such as recalling memories, summarizing experiences, or strategic planning -- based on temporal patterns and historical context. This functional approach allows cognitive modules to be dynamically added or removed without structural reengineering. Meanwhile, we propose a meta-learning strategy for continual policy adaptation, where the scheduler optimizes its cognitive strategy over time using historical interaction logs. Evaluation results demonstrate that our approach effectively learns to schedule cognitive activities based on historical data and can autonomously integrate new thinking modules.
-
Abstract
This paper examines memory mechanisms in Large Language Models (LLMs), emphasizing their importance for context-rich responses, reduced hallucinations, and improved efficiency. It categorizes memory into sensory, short-term, and long-term, with sensory memory corresponding to input prompts, short-term memory processing immediate context, and long-term memory implemented via external databases or structures. The text-based memory section covers acquisition (selection and summarization), management (updating, accessing, storing, and resolving conflicts), and utilization (full-text search, SQL queries, semantic search). The KV cache-based memory section discusses selection methods (regularity-based summarization, score-based approaches, special token embeddings) and compression techniques (low-rank compression, KV merging, multimodal compression), along with management strategies like offloading and shared attention mechanisms. Parameter-based memory methods (LoRA, TTT, MoE) transform memories into model parameters to enhance efficiency, while hidden-state-based memory approaches (chunk mechanisms, recurrent transformers, Mamba model) improve long-text processing by combining RNN hidden states with current methods. Overall, the paper offers a comprehensive analysis of LLM memory mechanisms, highlighting their significance and future research directions.
-
Abstract
Large Language Models (LLMs) have demonstrated remarkable prowess in generating contextually coherent responses, yet their fixed context windows pose fundamental challenges for maintaining consistency over prolonged multi-session dialogues. We introduce Mem0, a scalable memory-centric architecture that addresses this issue by dynamically extracting, consolidating, and retrieving salient information from ongoing conversations. Building on this foundation, we further propose an enhanced variant that leverages graph-based memory representations to capture complex relational structures among conversational elements. Through comprehensive evaluations on LOCOMO benchmark, we systematically compare our approaches against six baseline categories: (i) established memory-augmented systems, (ii) retrieval-augmented generation (RAG) with varying chunk sizes and k-values, (iii) a full-context approach that processes the entire conversation history, (iv) an open-source memory solution, (v) a proprietary model system, and (vi) a dedicated memory management platform. Empirical results show that our methods consistently outperform all existing memory systems across four question categories: single-hop, temporal, multi-hop, and open-domain. Notably, Mem0 achieves 26% relative improvements in the LLM-as-a-Judge metric over OpenAI, while Mem0 with graph memory achieves around 2% higher overall score than the base configuration. Beyond accuracy gains, we also markedly reduce computational overhead compared to full-context method. In particular, Mem0 attains a 91% lower p95 latency and saves more than 90% token cost, offering a compelling balance between advanced reasoning capabilities and practical deployment constraints. Our findings highlight critical role of structured, persistent memory mechanisms for long-term conversational coherence, paving the way for more reliable and efficient LLM-driven AI agents.
-
Abstract
Large Language Models (LLMs) have made significant progress in open-ended dialogue, yet their inability to retain and retrieve relevant information from long-term interactions limits their effectiveness in applications requiring sustained personalization. External memory mechanisms have been proposed to address this limitation, enabling LLMs to maintain conversational continuity. However, existing approaches struggle with two key challenges. First, rigid memory granularity fails to capture the natural semantic structure of conversations, leading to fragmented and incomplete representations. Second, fixed retrieval mechanisms cannot adapt to diverse dialogue contexts and user interaction patterns. In this work, we propose Reflective Memory Management (RMM), a novel mechanism for long-term dialogue agents, integrating forward- and backward-looking reflections: (1) Prospective Reflection, which dynamically summarizes interactions across granularities-utterances, turns, and sessions-into a personalized memory bank for effective future retrieval, and (2) Retrospective Reflection, which iteratively refines the retrieval in an online reinforcement learning (RL) manner based on LLMs' cited evidence. Experiments show that RMM demonstrates consistent improvement across various metrics and benchmarks. For example, RMM shows more than 10% accuracy improvement over the baseline without memory management on the LongMemEval dataset.
-
Abstract
LLM-based agents represent a paradigm shift in AI, enabling autonomous systems to plan, reason, and use tools while interacting with dynamic environments. This paper provides the first comprehensive survey of evaluation methods for these increasingly capable agents. We analyze the field of agent evaluation across five perspectives: (1) Core LLM capabilities needed for agentic workflows, like planning, and tool use; (2) Application-specific benchmarks such as web and SWE agents; (3) Evaluation of generalist agents; (4) Analysis of agent benchmarks' core dimensions; and (5) Evaluation frameworks and tools for agent developers. Our analysis reveals current trends, including a shift toward more realistic, challenging evaluations with continuously updated benchmarks. We also identify critical gaps that future research must address, particularly in assessing cost-efficiency, safety, and robustness, and in developing fine-grained, scalable evaluation methods.
-
Abstract
Agents powered by large language models (LLMs) have demonstrated strong capabilities in a wide range of complex, real-world applications. However, LLM agents with a compromised memory bank may easily produce harmful outputs when the past records retrieved for demonstration are malicious. In this paper, we propose a novel Memory INJection Attack, MINJA, without assuming that the attacker can directly modify the memory bank of the agent. The attacker injects malicious records into the memory bank by only interacting with the agent via queries and output observations. These malicious records are designed to elicit a sequence of malicious reasoning steps corresponding to a different target query during the agent's execution of the victim user's query. Specifically, we introduce a sequence of bridging steps to link victim queries to the malicious reasoning steps. During the memory injection, we propose an indication prompt that guides the agent to autonomously generate similar bridging steps, with a progressive shortening strategy that gradually removes the indication prompt, such that the malicious record will be easily retrieved when processing later victim queries. Our extensive experiments across diverse agents demonstrate the effectiveness of MINJA in compromising agent memory. With minimal requirements for execution, MINJA enables any user to influence agent memory, highlighting the risk.
-
Abstract
Human interaction with the external world fundamentally involves the exchange of personal memory, whether with other individuals, websites, applications, or, in the future, AI agents. A significant portion of this interaction is redundant, requiring users to repeatedly provide the same information across different contexts. Existing solutions, such as browser-stored credentials, autofill mechanisms, and unified authentication systems, have aimed to mitigate this redundancy by serving as intermediaries that store and retrieve commonly used user data. The advent of large language models (LLMs) presents an opportunity to redefine memory management through an AI-native paradigm: SECOND ME. SECOND ME acts as an intelligent, persistent memory offload system that retains, organizes, and dynamically utilizes user-specific knowledge. By serving as an intermediary in user interactions, it can autonomously generate context-aware responses, prefill required information, and facilitate seamless communication with external systems, significantly reducing cognitive load and interaction friction. Unlike traditional memory storage solutions, SECOND ME extends beyond static data retention by leveraging LLM-based memory parameterization. This enables structured organization, contextual reasoning, and adaptive knowledge retrieval, facilitating a more systematic and intelligent approach to memory management. As AI-driven personal agents like SECOND ME become increasingly integrated into digital ecosystems, SECOND ME further represents a critical step toward augmenting human-world interaction with persistent, contextually aware, and self-optimizing memory systems. We have open-sourced the fully localizable deployment system at GitHub: https://github.com/Mindverse/Second-Me.
-
Abstract
Long-term, open-domain dialogue capabilities are essential for chatbots aiming to recall past interactions and demonstrate emotional intelligence (EI). Yet, most existing research relies on synthetic, LLM-generated data, leaving open questions about real-world conversational patterns. To address this gap, we introduce REALTALK, a 21-day corpus of authentic messaging app dialogues, providing a direct benchmark against genuine human interactions. We first conduct a dataset analysis, focusing on EI attributes and persona consistency to understand the unique challenges posed by real-world dialogues. By comparing with LLM-generated conversations, we highlight key differences, including diverse emotional expressions and variations in persona stability that synthetic dialogues often fail to capture. Building on these insights, we introduce two benchmark tasks: (1) persona simulation where a model continues a conversation on behalf of a specific user given prior dialogue context; and (2) memory probing where a model answers targeted questions requiring long-term memory of past interactions. Our findings reveal that models struggle to simulate a user solely from dialogue history, while fine-tuning on specific user chats improves persona emulation. Additionally, existing models face significant challenges in recalling and leveraging long-term context within real-world conversations.
-
Abstract
Retrieval-augmented generation (RAG) enhances large language models (LLMs) for domain-specific question-answering (QA) tasks by leveraging external knowledge sources. However, traditional RAG systems primarily focus on relevance-based retrieval and often struggle with redundancy, especially when reasoning requires connecting information from multiple sources. This paper introduces Vendi-RAG, a framework based on an iterative process that jointly optimizes retrieval diversity and answer quality. This joint optimization leads to significantly higher accuracy for multi-hop QA tasks. Vendi-RAG leverages the Vendi Score (VS), a flexible similarity-based diversity metric, to promote semantic diversity in document retrieval. It then uses an LLM judge that evaluates candidate answers, generated after a reasoning step, and outputs a score that the retriever uses to balance relevance and diversity among the retrieved documents during each iteration. Experiments on three challenging datasets -- HotpotQA, MuSiQue, and 2WikiMultiHopQA -- demonstrate Vendi-RAG's effectiveness in multi-hop reasoning tasks. The framework achieves significant accuracy improvements over traditional single-step and multi-step RAG approaches, with accuracy increases reaching up to +4.2% on HotpotQA, +4.1% on 2WikiMultiHopQA, and +1.3% on MuSiQue compared to Adaptive-RAG, the current best baseline. The benefits of Vendi-RAG are even more pronounced as the number of retrieved documents increases. Finally, we evaluated Vendi-RAG across different LLM backbones, including GPT-3.5, GPT-4, and GPT-4o-mini, and observed consistent improvements, demonstrating that the framework's advantages are model-agnostic.
-
Abstract
Large Language Model (LLM) agents have become increasingly prevalent across various real-world applications. They enhance decision-making by storing private user-agent interactions in the memory module for demonstrations, introducing new privacy risks for LLM agents. In this work, we systematically investigate the vulnerability of LLM agents to our proposed Memory EXTRaction Attack (MEXTRA) under a black-box setting. To extract private information from memory, we propose an effective attacking prompt design and an automated prompt generation method based on different levels of knowledge about the LLM agent. Experiments on two representative agents demonstrate the effectiveness of MEXTRA. Moreover, we explore key factors influencing memory leakage from both the agent designer's and the attacker's perspectives. Our findings highlight the urgent need for effective memory safeguards in LLM agent design and deployment.
-
Abstract
While large language model (LLM) agents can effectively use external tools for complex real-world tasks, they require memory systems to leverage historical experiences. Current memory systems enable basic storage and retrieval but lack sophisticated memory organization, despite recent attempts to incorporate graph databases. Moreover, these systems' fixed operations and structures limit their adaptability across diverse tasks. To address this limitation, this paper proposes a novel agentic memory system for LLM agents that can dynamically organize memories in an agentic way. Following the basic principles of the Zettelkasten method, we designed our memory system to create interconnected knowledge networks through dynamic indexing and linking. When a new memory is added, we generate a comprehensive note containing multiple structured attributes, including contextual descriptions, keywords, and tags. The system then analyzes historical memories to identify relevant connections, establishing links where meaningful similarities exist. Additionally, this process enables memory evolution - as new memories are integrated, they can trigger updates to the contextual representations and attributes of existing historical memories, allowing the memory network to continuously refine its understanding. Our approach combines the structured organization principles of Zettelkasten with the flexibility of agent-driven decision making, allowing for more adaptive and context-aware memory management. Empirical experiments on six foundation models show superior improvement against existing SOTA baselines. The source code for evaluating performance is available at https://github.com/WujiangXu/A-mem, while the source code of the agentic memory system is available at https://github.com/WujiangXu/A-mem-sys.
-
Abstract
As Large Language Models (LLMs) evolve from text-completion tools into fully fledged agents operating in dynamic environments, they must address the challenge of continually learning and retaining long-term knowledge. Many biological systems solve these challenges with episodic memory, which supports single-shot learning of instance-specific contexts. Inspired by this, we present an episodic memory framework for LLM agents, centered around five key properties of episodic memory that underlie adaptive and context-sensitive behavior. With various research efforts already partially covering these properties, this position paper argues that now is the right time for an explicit, integrated focus on episodic memory to catalyze the development of long-term agents. To this end, we outline a roadmap that unites several research directions under the goal to support all five properties of episodic memory for more efficient long-term LLM agents.
-
Abstract
Log-Structured Merge-tree-based Key-Value Store (LSM-KVS) is a foundational storage engine serving diverse modern workloads, systems, and applications. To suit varying use cases, LSM-KVS allows a vast configuration space that controls core parameters like compaction, flush, and cache sizes, each consuming a shared pool of CPU, Memory, and Storage resources. Navigating the LSM-KVS configuration space necessitates knowledge of the impact of each configuration on the expected workload and underlying hardware. Beyond expensive and time-intensive human-expert-based tuning, existing LSM-KVS tuning solutions focus on tuning with specific workload expectations while limited to a narrow subset of parameters. This paper introduces ELMo-Tune-V2, a framework that integrates Large Language Models (LLMs) at its foundation to demonstrate the potential of applying modern LLMs in data system optimization problems. ELMo-Tune-V2 leverages the contextual reasoning, cross-domain, and generative capabilities of LLMs to perform 1) self-navigated characterization and modeling of LSM-KVS workloads, 2) automatic tuning across a broad parameter space using cross-domain knowledge, and 3) real-time dynamic configuration adjustments for LSM-KVS. ELMo-Tune-V2 integrates three innovations: LLM-based workload synthesis for adaptive benchmark generation, feedback-driven iterative fine-tuning for configuration refinement, and real-time tuning to handle evolving workloads. Through detailed evaluation using RocksDB under several real-world applications across diverse scenarios, ELMo-Tune-V2 achieves performance improvements up to ~14X our YCSB benchmarks compared against default RocksDB configurations, and our end-to-end tests with upper-level applications, NebulaGraph and Kvrocks, demonstrate performance gains of 34% and 26%, respectively.
-
Abstract
We introduce Zep, a novel memory layer service for AI agents that outperforms the current state-of-the-art system, MemGPT, in the Deep Memory Retrieval (DMR) benchmark. Additionally, Zep excels in more comprehensive and challenging evaluations than DMR that better reflect real-world enterprise use cases. While existing retrieval-augmented generation (RAG) frameworks for large language model (LLM)-based agents are limited to static document retrieval, enterprise applications demand dynamic knowledge integration from diverse sources including ongoing conversations and business data. Zep addresses this fundamental limitation through its core component Graphiti -- a temporally-aware knowledge graph engine that dynamically synthesizes both unstructured conversational data and structured business data while maintaining historical relationships. In the DMR benchmark, which the MemGPT team established as their primary evaluation metric, Zep demonstrates superior performance (94.8% vs 93.4%). Beyond DMR, Zep's capabilities are further validated through the more challenging LongMemEval benchmark, which better reflects enterprise use cases through complex temporal reasoning tasks. In this evaluation, Zep achieves substantial results with accuracy improvements of up to 18.5% while simultaneously reducing response latency by 90% compared to baseline implementations. These results are particularly pronounced in enterprise-critical tasks such as cross-session information synthesis and long-term context maintenance, demonstrating Zep's effectiveness for deployment in real-world applications.
-
Abstract
Most current AI models have little ability to store and later retrieve a record or representation of what they do. In human cognition, episodic memories play an important role in both recall of the past as well as planning for the future. The ability to form and use episodic memories would similarly enable a broad range of improved capabilities in an AI agent that interacts with and takes actions in the world. Researchers have begun directing more attention to developing memory abilities in AI models. It is therefore likely that models with such capability will be become widespread in the near future. This could in some ways contribute to making such AI agents safer by enabling users to better monitor, understand, and control their actions. However, as a new capability with wide applications, we argue that it will also introduce significant new risks that researchers should begin to study and address. We outline these risks and benefits and propose four principles to guide the development of episodic memory capabilities so that these will enhance, rather than undermine, the effort to keep AI safe and trustworthy.
-
Abstract
Lifelong learning, also known as continual or incremental learning, is a crucial component for advancing Artificial General Intelligence (AGI) by enabling systems to continuously adapt in dynamic environments. While large language models (LLMs) have demonstrated impressive capabilities in natural language processing, existing LLM agents are typically designed for static systems and lack the ability to adapt over time in response to new challenges. This survey is the first to systematically summarize the potential techniques for incorporating lifelong learning into LLM-based agents. We categorize the core components of these agents into three modules: the perception module for multimodal input integration, the memory module for storing and retrieving evolving knowledge, and the action module for grounded interactions with the dynamic environment. We highlight how these pillars collectively enable continuous adaptation, mitigate catastrophic forgetting, and improve long-term performance. This survey provides a roadmap for researchers and practitioners working to develop lifelong learning capabilities in LLM agents, offering insights into emerging trends, evaluation metrics, and application scenarios. Relevant literature and resources are available at \href{this url}{https://github.com/qianlima-lab/awesome-lifelong-llm-agent}.
-
Abstract
Synthetic data generation with Large Language Models is a promising paradigm for augmenting natural data over a nearly infinite range of tasks. Given this variety, direct comparisons among synthetic data generation algorithms are scarce, making it difficult to understand where improvement comes from and what bottlenecks exist. We propose to evaluate algorithms via the makeup of synthetic data generated by each algorithm in terms of data quality, diversity, and complexity. We choose these three characteristics for their significance in open-ended processes and the impact each has on the capabilities of downstream models. We find quality to be essential for in-distribution model generalization, diversity to be essential for out-of-distribution generalization, and complexity to be beneficial for both. Further, we emphasize the existence of Quality-Diversity trade-offs in training data and the downstream effects on model performance. We then examine the effect of various components in the synthetic data pipeline on each data characteristic. This examination allows us to taxonomize and compare synthetic data generation algorithms through the components they utilize and the resulting effects on data QDC composition. This analysis extends into a discussion on the importance of balancing QDC in synthetic data for efficient reinforcement learning and self-improvement algorithms. Analogous to the QD trade-offs in training data, often there exist trade-offs between model output quality and output diversity which impact the composition of synthetic data. We observe that many models are currently evaluated and optimized only for output quality, thereby limiting output diversity and the potential for self-improvement. We argue that balancing these trade-offs is essential to the development of future self-improvement algorithms and highlight a number of works making progress in this direction.
-
Abstract
Modern AI agents, driven by advances in large foundation models, promise to enhance our productivity and transform our lives by augmenting our knowledge and capabilities. To achieve this vision, AI agents must effectively plan, perform multi-step reasoning and actions, respond to novel observations, and recover from errors, to successfully complete complex tasks across a wide range of scenarios. In this work, we introduce Magentic-One, a high-performing open-source agentic system for solving such tasks. Magentic-One uses a multi-agent architecture where a lead agent, the Orchestrator, plans, tracks progress, and re-plans to recover from errors. Throughout task execution, the Orchestrator directs other specialized agents to perform tasks as needed, such as operating a web browser, navigating local files, or writing and executing Python code. We show that Magentic-One achieves statistically competitive performance to the state-of-the-art on three diverse and challenging agentic benchmarks: GAIA, AssistantBench, and WebArena. Magentic-One achieves these results without modification to core agent capabilities or to how they collaborate, demonstrating progress towards generalist agentic systems. Moreover, Magentic-One's modular design allows agents to be added or removed from the team without additional prompt tuning or training, easing development and making it extensible to future scenarios. We provide an open-source implementation of Magentic-One, and we include AutoGenBench, a standalone tool for agentic evaluation. AutoGenBench provides built-in controls for repetition and isolation to run agentic benchmarks in a rigorous and contained manner -- which is important when agents' actions have side-effects. Magentic-One, AutoGenBench and detailed empirical performance evaluations of Magentic-One, including ablations and error analysis are available at https://aka.ms/magentic-one
-
Abstract
Vector Symbolic Architectures (VSAs) are one approach to developing Neuro-symbolic AI, where two vectors in $\mathbb{R}^d$ are `bound' together to produce a new vector in the same space. VSAs support the commutativity and associativity of this binding operation, along with an inverse operation, allowing one to construct symbolic-style manipulations over real-valued vectors. Most VSAs were developed before deep learning and automatic differentiation became popular and instead focused on efficacy in hand-designed systems. In this work, we introduce the Hadamard-derived linear Binding (HLB), which is designed to have favorable computational efficiency, and efficacy in classic VSA tasks, and perform well in differentiable systems. Code is available at https://github.com/FutureComputing4AI/Hadamard-derived-Linear-Binding
-
Abstract
Recent large language model (LLM)-driven chat assistant systems have integrated memory components to track user-assistant chat histories, enabling more accurate and personalized responses. However, their long-term memory capabilities in sustained interactions remain underexplored. We introduce LongMemEval, a comprehensive benchmark designed to evaluate five core long-term memory abilities of chat assistants: information extraction, multi-session reasoning, temporal reasoning, knowledge updates, and abstention. With 500 meticulously curated questions embedded within freely scalable user-assistant chat histories, LongMemEval presents a significant challenge to existing long-term memory systems, with commercial chat assistants and long-context LLMs showing a 30% accuracy drop on memorizing information across sustained interactions. We then present a unified framework that breaks down the long-term memory design into three stages: indexing, retrieval, and reading. Built upon key experimental insights, we propose several memory design optimizations including session decomposition for value granularity, fact-augmented key expansion for indexing, and time-aware query expansion for refining the search scope. Extensive experiments show that these optimizations greatly improve both memory recall and downstream question answering on LongMemEval. Overall, our study provides valuable resources and guidance for advancing the long-term memory capabilities of LLM-based chat assistants, paving the way toward more personalized and reliable conversational AI. Our benchmark and code are publicly available at https://github.com/xiaowu0162/LongMemEval.
-
Abstract
Recent advancements in large language models have significantly improved their context windows, yet challenges in effective long-term memory management remain. We introduce MemTree, an algorithm that leverages a dynamic, tree-structured memory representation to optimize the organization, retrieval, and integration of information, akin to human cognitive schemas. MemTree organizes memory hierarchically, with each node encapsulating aggregated textual content, corresponding semantic embeddings, and varying abstraction levels across the tree's depths. Our algorithm dynamically adapts this memory structure by computing and comparing semantic embeddings of new and existing information to enrich the model's context-awareness. This approach allows MemTree to handle complex reasoning and extended interactions more effectively than traditional memory augmentation methods, which often rely on flat lookup tables. Evaluations on benchmarks for multi-turn dialogue understanding and document question answering show that MemTree significantly enhances performance in scenarios that demand structured memory management.
-
Abstract
Despite the potential of language model-based agents to solve real-world tasks such as web navigation, current methods still struggle with long-horizon tasks with complex action trajectories. In contrast, humans can flexibly solve complex tasks by learning reusable task workflows from past experiences and using them to guide future actions. To build agents that can similarly benefit from this process, we introduce Agent Workflow Memory (AWM), a method for inducing commonly reused routines, i.e., workflows, and selectively providing workflows to the agent to guide subsequent generations. AWM flexibly applies to both offline and online scenarios, where agents induce workflows from training examples beforehand or from test queries on the fly. We experiment on two major web navigation benchmarks -- Mind2Web and WebArena -- that collectively cover 1000+ tasks from 200+ domains across travel, shopping, and social media, among others. AWM substantially improves the baseline results by 24.6% and 51.1% relative success rate on Mind2Web and WebArena while reducing the number of steps taken to solve WebArena tasks successfully. Furthermore, online AWM robustly generalizes in cross-task, website, and domain evaluations, surpassing baselines from 8.9 to 14.0 absolute points as train-test task distribution gaps widen.
-
Abstract
Extraction and interpretation of intricate information from unstructured text data arising in financial applications, such as earnings call transcripts, present substantial challenges to large language models (LLMs) even using the current best practices to use Retrieval Augmented Generation (RAG) (referred to as VectorRAG techniques which utilize vector databases for information retrieval) due to challenges such as domain specific terminology and complex formats of the documents. We introduce a novel approach based on a combination, called HybridRAG, of the Knowledge Graphs (KGs) based RAG techniques (called GraphRAG) and VectorRAG techniques to enhance question-answer (Q&A) systems for information extraction from financial documents that is shown to be capable of generating accurate and contextually relevant answers. Using experiments on a set of financial earning call transcripts documents which come in the form of Q&A format, and hence provide a natural set of pairs of ground-truth Q&As, we show that HybridRAG which retrieves context from both vector database and KG outperforms both traditional VectorRAG and GraphRAG individually when evaluated at both the retrieval and generation stages in terms of retrieval accuracy and answer generation. The proposed technique has applications beyond the financial domain
-
Abstract
As language models (LMs) are widely utilized in personalized communication scenarios (e.g., sending emails, writing social media posts) and endowed with a certain level of agency, ensuring they act in accordance with the contextual privacy norms becomes increasingly critical. However, quantifying the privacy norm awareness of LMs and the emerging privacy risk in LM-mediated communication is challenging due to (1) the contextual and long-tailed nature of privacy-sensitive cases, and (2) the lack of evaluation approaches that capture realistic application scenarios. To address these challenges, we propose PrivacyLens, a novel framework designed to extend privacy-sensitive seeds into expressive vignettes and further into agent trajectories, enabling multi-level evaluation of privacy leakage in LM agents' actions. We instantiate PrivacyLens with a collection of privacy norms grounded in privacy literature and crowdsourced seeds. Using this dataset, we reveal a discrepancy between LM performance in answering probing questions and their actual behavior when executing user instructions in an agent setup. State-of-the-art LMs, like GPT-4 and Llama-3-70B, leak sensitive information in 25.68% and 38.69% of cases, even when prompted with privacy-enhancing instructions. We also demonstrate the dynamic nature of PrivacyLens by extending each seed into multiple trajectories to red-team LM privacy leakage risk. Dataset and code are available at https://github.com/SALT-NLP/PrivacyLens.
-
Abstract
Advancements in the capabilities of Large Language Models (LLMs) have created a promising foundation for developing autonomous agents. With the right tools, these agents could learn to solve tasks in new environments by accumulating and updating their knowledge. Current LLM-based agents process past experiences using a full history of observations, summarization, retrieval augmentation. However, these unstructured memory representations do not facilitate the reasoning and planning essential for complex decision-making. In our study, we introduce AriGraph, a novel method wherein the agent constructs and updates a memory graph that integrates semantic and episodic memories while exploring the environment. We demonstrate that our Ariadne LLM agent, consisting of the proposed memory architecture augmented with planning and decision-making, effectively handles complex tasks within interactive text game environments difficult even for human players. Results show that our approach markedly outperforms other established memory methods and strong RL baselines in a range of problems of varying complexity. Additionally, AriGraph demonstrates competitive performance compared to dedicated knowledge graph-based methods in static multi-hop question-answering.
-
Abstract
LLM agents have demonstrated remarkable performance across various applications, primarily due to their advanced capabilities in reasoning, utilizing external knowledge and tools, calling APIs, and executing actions to interact with environments. Current agents typically utilize a memory module or a retrieval-augmented generation (RAG) mechanism, retrieving past knowledge and instances with similar embeddings from knowledge bases to inform task planning and execution. However, the reliance on unverified knowledge bases raises significant concerns about their safety and trustworthiness. To uncover such vulnerabilities, we propose a novel red teaming approach AgentPoison, the first backdoor attack targeting generic and RAG-based LLM agents by poisoning their long-term memory or RAG knowledge base. In particular, we form the trigger generation process as a constrained optimization to optimize backdoor triggers by mapping the triggered instances to a unique embedding space, so as to ensure that whenever a user instruction contains the optimized backdoor trigger, the malicious demonstrations are retrieved from the poisoned memory or knowledge base with high probability. In the meantime, benign instructions without the trigger will still maintain normal performance. Unlike conventional backdoor attacks, AgentPoison requires no additional model training or fine-tuning, and the optimized backdoor trigger exhibits superior transferability, in-context coherence, and stealthiness. Extensive experiments demonstrate AgentPoison's effectiveness in attacking three types of real-world LLM agents: RAG-based autonomous driving agent, knowledge-intensive QA agent, and healthcare EHRAgent. On each agent, AgentPoison achieves an average attack success rate higher than 80% with minimal impact on benign performance (less than 1%) with a poison rate less than 0.1%.
-
Abstract
Commonly used caching policies, such as LRU (Least Recently Used) or LFU (Least Frequently Used), exhibit optimal performance only under specific traffic patterns. Even advanced machine learning-based methods, which detect patterns in historical request data, struggle when future requests deviate from past trends. Recently, a new class of policies has emerged that are robust to varying traffic patterns. These algorithms address an online optimization problem, enabling continuous adaptation to the context. They offer theoretical guarantees on the regret metric, which measures the performance gap between the online policy and the optimal static cache allocation in hindsight. However, the high computational complexity of these solutions hinders their practical adoption. In this study, we introduce a new variant of the gradient-based online caching policy that achieves groundbreaking logarithmic computational complexity relative to catalog size, while also providing regret guarantees. This advancement allows us to test the policy on large-scale, real-world traces featuring millions of requests and items - a significant achievement, as such scales have been beyond the reach of existing policies with regret guarantees. To the best of our knowledge, our experimental results demonstrate for the first time that the regret guarantees of gradient-based caching policies offer substantial benefits in practical scenarios.
-
Abstract
In order to thrive in hostile and ever-changing natural environments, mammalian brains evolved to store large amounts of knowledge about the world and continually integrate new information while avoiding catastrophic forgetting. Despite the impressive accomplishments, large language models (LLMs), even with retrieval-augmented generation (RAG), still struggle to efficiently and effectively integrate a large amount of new experiences after pre-training. In this work, we introduce HippoRAG, a novel retrieval framework inspired by the hippocampal indexing theory of human long-term memory to enable deeper and more efficient knowledge integration over new experiences. HippoRAG synergistically orchestrates LLMs, knowledge graphs, and the Personalized PageRank algorithm to mimic the different roles of neocortex and hippocampus in human memory. We compare HippoRAG with existing RAG methods on multi-hop question answering and show that our method outperforms the state-of-the-art methods remarkably, by up to 20%. Single-step retrieval with HippoRAG achieves comparable or better performance than iterative retrieval like IRCoT while being 10-30 times cheaper and 6-13 times faster, and integrating HippoRAG into IRCoT brings further substantial gains. Finally, we show that our method can tackle new types of scenarios that are out of reach of existing methods. Code and data are available at https://github.com/OSU-NLP-Group/HippoRAG.
-
Abstract
Existing works on long-term open-domain dialogues focus on evaluating model responses within contexts spanning no more than five chat sessions. Despite advancements in long-context large language models (LLMs) and retrieval augmented generation (RAG) techniques, their efficacy in very long-term dialogues remains unexplored. To address this research gap, we introduce a machine-human pipeline to generate high-quality, very long-term dialogues by leveraging LLM-based agent architectures and grounding their dialogues on personas and temporal event graphs. Moreover, we equip each agent with the capability of sharing and reacting to images. The generated conversations are verified and edited by human annotators for long-range consistency and grounding to the event graphs. Using this pipeline, we collect LoCoMo, a dataset of very long-term conversations, each encompassing 300 turns and 9K tokens on avg., over up to 35 sessions. Based on LoCoMo, we present a comprehensive evaluation benchmark to measure long-term memory in models, encompassing question answering, event summarization, and multi-modal dialogue generation tasks. Our experimental results indicate that LLMs exhibit challenges in understanding lengthy conversations and comprehending long-range temporal and causal dynamics within dialogues. Employing strategies like long-context LLMs or RAG can offer improvements but these models still substantially lag behind human performance.
-
Abstract
Large language models (LLMs) have revolutionized AI, but are constrained by limited context windows, hindering their utility in tasks like extended conversations and document analysis. To enable using context beyond limited context windows, we propose virtual context management, a technique drawing inspiration from hierarchical memory systems in traditional operating systems that provide the appearance of large memory resources through data movement between fast and slow memory. Using this technique, we introduce MemGPT (Memory-GPT), a system that intelligently manages different memory tiers in order to effectively provide extended context within the LLM's limited context window, and utilizes interrupts to manage control flow between itself and the user. We evaluate our OS-inspired design in two domains where the limited context windows of modern LLMs severely handicaps their performance: document analysis, where MemGPT is able to analyze large documents that far exceed the underlying LLM's context window, and multi-session chat, where MemGPT can create conversational agents that remember, reflect, and evolve dynamically through long-term interactions with their users. We release MemGPT code and data for our experiments at https://memgpt.ai.
-
Abstract
Recent efforts have augmented large language models (LLMs) with external resources (e.g., the Internet) or internal control flows (e.g., prompt chaining) for tasks requiring grounding or reasoning, leading to a new class of language agents. While these agents have achieved substantial empirical success, we lack a systematic framework to organize existing agents and plan future developments. In this paper, we draw on the rich history of cognitive science and symbolic artificial intelligence to propose Cognitive Architectures for Language Agents (CoALA). CoALA describes a language agent with modular memory components, a structured action space to interact with internal memory and external environments, and a generalized decision-making process to choose actions. We use CoALA to retrospectively survey and organize a large body of recent work, and prospectively identify actionable directions towards more capable agents. Taken together, CoALA contextualizes today's language agents within the broader history of AI and outlines a path towards language-based general intelligence.
-
Abstract
Hyperdimensional computing (HDC) is an emerging computing paradigm that imitates the brain's structure to offer a powerful and efficient processing and learning model. In HDC, the data are encoded with long vectors, called hypervectors, typically with a length of 1K to 10K. The literature provides several encoding techniques to generate orthogonal or correlated hypervectors, depending on the intended application. The existing surveys in the literature often focus on the overall aspects of HDC systems, including system inputs, primary computations, and final outputs. However, this study takes a more specific approach. It zeroes in on the HDC system input and the generation of hypervectors, directly influencing the hypervector encoding process. This survey brings together various methods for hypervector generation from different studies and explores the limitations, challenges, and potential benefits they entail. Through a comprehensive exploration of this survey, readers will acquire a profound understanding of various encoding types in HDC and gain insights into the intricate process of hypervector generation for diverse applications.
-
Abstract
Metrics for set similarity are a core aspect of several data mining tasks. To remove duplicate results in a Web search, for example, a common approach looks at the Jaccard index between all pairs of pages. In social network analysis, a much-celebrated metric is the Adamic-Adar index, widely used to compare node neighborhood sets in the important problem of predicting links. However, with the increasing amount of data to be processed, calculating the exact similarity between all pairs can be intractable. The challenge of working at this scale has motivated research into efficient estimators for set similarity metrics. The two most popular estimators, MinHash and SimHash, are indeed used in applications such as document deduplication and recommender systems where large volumes of data need to be processed. Given the importance of these tasks, the demand for advancing estimators is evident. We propose DotHash, an unbiased estimator for the intersection size of two sets. DotHash can be used to estimate the Jaccard index and, to the best of our knowledge, is the first method that can also estimate the Adamic-Adar index and a family of related metrics. We formally define this family of metrics, provide theoretical bounds on the probability of estimate errors, and analyze its empirical performance. Our experimental results indicate that DotHash is more accurate than the other estimators in link prediction and detecting duplicate documents with the same complexity and similar comparison time.
-
Abstract
Both Minsky's "society of mind" and Schmidhuber's "learning to think" inspire diverse societies of large multimodal neural networks (NNs) that solve problems by interviewing each other in a "mindstorm." Recent implementations of NN-based societies of minds consist of large language models (LLMs) and other NN-based experts communicating through a natural language interface. In doing so, they overcome the limitations of single LLMs, improving multimodal zero-shot reasoning. In these natural language-based societies of mind (NLSOMs), new agents -- all communicating through the same universal symbolic language -- are easily added in a modular fashion. To demonstrate the power of NLSOMs, we assemble and experiment with several of them (having up to 129 members), leveraging mindstorms in them to solve some practical AI tasks: visual question answering, image captioning, text-to-image synthesis, 3D generation, egocentric retrieval, embodied AI, and general language-based task solving. We view this as a starting point towards much larger NLSOMs with billions of agents-some of which may be humans. And with this emergence of great societies of heterogeneous minds, many new research questions have suddenly become paramount to the future of artificial intelligence. What should be the social structure of an NLSOM? What would be the (dis)advantages of having a monarchical rather than a democratic structure? How can principles of NN economies be used to maximize the total reward of a reinforcement learning NLSOM? In this work, we identify, discuss, and try to answer some of these questions.
-
Abstract
Revolutionary advancements in Large Language Models have drastically reshaped our interactions with artificial intelligence systems. Despite this, a notable hindrance remains-the deficiency of a long-term memory mechanism within these models. This shortfall becomes increasingly evident in situations demanding sustained interaction, such as personal companion systems and psychological counseling. Therefore, we propose MemoryBank, a novel memory mechanism tailored for LLMs. MemoryBank enables the models to summon relevant memories, continually evolve through continuous memory updates, comprehend, and adapt to a user personality by synthesizing information from past interactions. To mimic anthropomorphic behaviors and selectively preserve memory, MemoryBank incorporates a memory updating mechanism, inspired by the Ebbinghaus Forgetting Curve theory, which permits the AI to forget and reinforce memory based on time elapsed and the relative significance of the memory, thereby offering a human-like memory mechanism. MemoryBank is versatile in accommodating both closed-source models like ChatGPT and open-source models like ChatGLM. We exemplify application of MemoryBank through the creation of an LLM-based chatbot named SiliconFriend in a long-term AI Companion scenario. Further tuned with psychological dialogs, SiliconFriend displays heightened empathy in its interactions. Experiment involves both qualitative analysis with real-world user dialogs and quantitative analysis with simulated dialogs. In the latter, ChatGPT acts as users with diverse characteristics and generates long-term dialog contexts covering a wide array of topics. The results of our analysis reveal that SiliconFriend, equipped with MemoryBank, exhibits a strong capability for long-term companionship as it can provide emphatic response, recall relevant memories and understand user personality.
-
Abstract
We introduce Voyager, the first LLM-powered embodied lifelong learning agent in Minecraft that continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager consists of three key components: 1) an automatic curriculum that maximizes exploration, 2) an ever-growing skill library of executable code for storing and retrieving complex behaviors, and 3) a new iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program improvement. Voyager interacts with GPT-4 via blackbox queries, which bypasses the need for model parameter fine-tuning. The skills developed by Voyager are temporally extended, interpretable, and compositional, which compounds the agent's abilities rapidly and alleviates catastrophic forgetting. Empirically, Voyager shows strong in-context lifelong learning capability and exhibits exceptional proficiency in playing Minecraft. It obtains 3.3x more unique items, travels 2.3x longer distances, and unlocks key tech tree milestones up to 15.3x faster than prior SOTA. Voyager is able to utilize the learned skill library in a new Minecraft world to solve novel tasks from scratch, while other techniques struggle to generalize. We open-source our full codebase and prompts at https://voyager.minedojo.org/.
-
Abstract
Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior.
-
Abstract
Large language models (LLMs) have been increasingly used to interact with external environments (e.g., games, compilers, APIs) as goal-driven agents. However, it remains challenging for these language agents to quickly and efficiently learn from trial-and-error as traditional reinforcement learning methods require extensive training samples and expensive model fine-tuning. We propose Reflexion, a novel framework to reinforce language agents not by updating weights, but instead through linguistic feedback. Concretely, Reflexion agents verbally reflect on task feedback signals, then maintain their own reflective text in an episodic memory buffer to induce better decision-making in subsequent trials. Reflexion is flexible enough to incorporate various types (scalar values or free-form language) and sources (external or internally simulated) of feedback signals, and obtains significant improvements over a baseline agent across diverse tasks (sequential decision-making, coding, language reasoning). For example, Reflexion achieves a 91% pass@1 accuracy on the HumanEval coding benchmark, surpassing the previous state-of-the-art GPT-4 that achieves 80%. We also conduct ablation and analysis studies using different feedback signals, feedback incorporation methods, and agent types, and provide insights into how they affect performance.
-
Abstract
Large Language Models (LLMs) are increasingly being integrated into various applications. The functionalities of recent LLMs can be flexibly modulated via natural language prompts. This renders them susceptible to targeted adversarial prompting, e.g., Prompt Injection (PI) attacks enable attackers to override original instructions and employed controls. So far, it was assumed that the user is directly prompting the LLM. But, what if it is not the user prompting? We argue that LLM-Integrated Applications blur the line between data and instructions. We reveal new attack vectors, using Indirect Prompt Injection, that enable adversaries to remotely (without a direct interface) exploit LLM-integrated applications by strategically injecting prompts into data likely to be retrieved. We derive a comprehensive taxonomy from a computer security perspective to systematically investigate impacts and vulnerabilities, including data theft, worming, information ecosystem contamination, and other novel security risks. We demonstrate our attacks' practical viability against both real-world systems, such as Bing's GPT-4 powered Chat and code-completion engines, and synthetic applications built on GPT-4. We show how processing retrieved prompts can act as arbitrary code execution, manipulate the application's functionality, and control how and if other APIs are called. Despite the increasing integration and reliance on LLMs, effective mitigations of these emerging threats are currently lacking. By raising awareness of these vulnerabilities and providing key insights into their implications, we aim to promote the safe and responsible deployment of these powerful models and the development of robust defenses that protect users and systems from potential attacks.
-
Abstract
Social insects such as ants communicate via pheromones which allows them to coordinate their activity and solve complex tasks as a swarm, e.g. foraging for food. This behavior was shaped through evolutionary processes. In computational models, self-coordination in swarms has been implemented using probabilistic or simple action rules to shape the decision of each agent and the collective behavior. However, manual tuned decision rules may limit the behavior of the swarm. In this work we investigate the emergence of self-coordination and communication in evolved swarms without defining any explicit rule. We evolve a swarm of agents representing an ant colony. We use an evolutionary algorithm to optimize a spiking neural network (SNN) which serves as an artificial brain to control the behavior of each agent. The goal of the evolved colony is to find optimal ways to forage for food and return it to the nest in the shortest amount of time. In the evolutionary phase, the ants are able to learn to collaborate by depositing pheromone near food piles and near the nest to guide other ants. The pheromone usage is not manually encoded into the network; instead, this behavior is established through the optimization procedure. We observe that pheromone-based communication enables the ants to perform better in comparison to colonies where communication via pheromone did not emerge. We assess the foraging performance by comparing the SNN based model to a rule based system. Our results show that the SNN based model can efficiently complete the foraging task in a short amount of time. Our approach illustrates self coordination via pheromone emerges as a result of the network optimization. This work serves as a proof of concept for the possibility of creating complex applications utilizing SNNs as underlying architectures for multi-agent interactions where communication and self-coordination is desired.
-
Abstract
It is challenging to determine whether datasets are findable, accessible, interoperable, and reusable (FAIR) because the FAIR Guiding Principles refer to highly idiosyncratic criteria regarding the metadata used to annotate datasets. Specifically, the FAIR principles require metadata to be "rich" and to adhere to "domain-relevant" community standards. Scientific communities should be able to define their own machine-actionable templates for metadata that encode these "rich," discipline-specific elements. We have explored this template-based approach in the context of two software systems. One system is the CEDAR Workbench, which investigators use to author new metadata. The other is the FAIRware Workbench, which evaluates the metadata of archived datasets for their adherence to community standards. Benefits accrue when templates for metadata become central elements in an ecosystem of tools to manage online datasets—both because the templates serve as a community reference for what constitutes FAIR data, and because they embody that perspective in a form that can be distributed among a variety of software applications to assist with data stewardship and data sharing.
-
Abstract
Remembering important information from the past and continuing to talk about it in the present are crucial in long-term conversations. However, previous literature does not deal with cases where the memorized information is outdated, which may cause confusion in later conversations. To address this issue, we present a novel task and a corresponding dataset of memory management in long-term conversations, in which bots keep track of and bring up the latest information about users while conversing through multiple sessions. In order to support more precise and interpretable memory, we represent memory as unstructured text descriptions of key information and propose a new mechanism of memory management that selectively eliminates invalidated or redundant information. Experimental results show that our approach outperforms the baselines that leave the stored memory unchanged in terms of engagingness and humanness, with larger performance gap especially in the later sessions.
-
Abstract
Provenance encodes information that connects datasets, their generation workflows, and associated metadata (e.g., who or when executed a query). As such, it is instrumental for a wide range of critical governance applications (e.g., observability and auditing). Unfortunately, in the context of database systems, extracting coarse-grained provenance is a long-standing problem due to the complexity and sheer volume of database workflows. Provenance extraction from query event logs has been recently proposed as favorable because, in principle, can result in meaningful provenance graphs for provenance applications. Current approaches, however, (a) add substantial overhead to the database and provenance extraction workflows and (b)~extract provenance that is noisy, omits query execution dependencies, and is not rich enough for upstream applications. To address these problems, we introduce OneProvenance: an efficient provenance extraction system from query event logs. OneProvenance addresses the unique challenges of log-based extraction by (a)~identifying query execution dependencies through efficient log analysis, (b) extracting provenance through novel event transformations that account for query dependencies, and (c)~introducing effective filtering optimizations. Our thorough experimental analysis shows that OneProvenance can improve extraction by up to ~18X compared to state-of-the-art baselines; our optimizations reduce the extraction noise and optimize performance even further. OneProvenance is deployed at scale by Microsoft Purview and actively supports customer provenance extraction needs (https://bit.ly/3N2JVGF).
-
Abstract
Fully-parametric language models generally require a huge number of model parameters to store the necessary knowledge for solving multiple natural language tasks in zero/few-shot settings. In addition, it is hard to adapt to the evolving world knowledge without the costly model re-training. In this paper, we develop a novel semi-parametric language model architecture, Knowledge-in-Context (KiC), which empowers a parametric text-to-text language model with a knowledge-rich external memory. Specifically, the external memory contains six different types of knowledge: entity, dictionary, commonsense, event, script, and causality knowledge. For each input instance, the KiC model adaptively selects a knowledge type and retrieves the most helpful pieces of knowledge. The input instance along with its knowledge augmentation is fed into a text-to-text model (e.g., T5) to generate the output answer, where both the input and the output are in natural language forms after prompting. Interestingly, we find that KiC can be identified as a special mixture-of-experts (MoE) model, where the knowledge selector plays the role of a router that is used to determine the sequence-to-expert assignment in MoE. This key observation inspires us to develop a novel algorithm for training KiC with an instance-adaptive knowledge selector. As a knowledge-rich semi-parametric language model, KiC only needs a much smaller parametric part to achieve superior zero-shot performance on unseen tasks. By evaluating on 40+ different tasks, we show that KiC_Large with 770M parameters easily outperforms large language models (LMs) that are 4-39x larger by a large margin. We also demonstrate that KiC exhibits emergent abilities at a much smaller model scale compared to the fully-parametric models.
-
Abstract
While language Models store a massive amount of world knowledge implicitly in their parameters, even very large models often fail to encode information about rare entities and events, while incurring huge computational costs. Recently, retrieval-augmented models, such as REALM, RAG, and RETRO, have incorporated world knowledge into language generation by leveraging an external non-parametric index and have demonstrated impressive performance with constrained model sizes. However, these methods are restricted to retrieving only textual knowledge, neglecting the ubiquitous amount of knowledge in other modalities like images -- much of which contains information not covered by any text. To address this limitation, we propose the first Multimodal Retrieval-Augmented Transformer (MuRAG), which accesses an external non-parametric multimodal memory to augment language generation. MuRAG is pre-trained with a mixture of large-scale image-text and text-only corpora using a joint contrastive and generative loss. We perform experiments on two different datasets that require retrieving and reasoning over both images and text to answer a given query: WebQA, and MultimodalQA. Our results show that MuRAG achieves state-of-the-art accuracy, outperforming existing models by 10-20\% absolute on both datasets and under both distractor and full-wiki settings.
-
Abstract
Today, computer systems hold large amounts of personal data. Yet while such an abundance of data allows breakthroughs in artificial intelligence, and especially machine learning (ML), its existence can be a threat to user privacy, and it can weaken the bonds of trust between humans and AI. Recent regulations now require that, on request, private information about a user must be removed from both computer systems and from ML models, i.e. ``the right to be forgotten''). While removing data from back-end databases should be straightforward, it is not sufficient in the AI context as ML models often `remember' the old data. Contemporary adversarial attacks on trained models have proven that we can learn whether an instance or an attribute belonged to the training data. This phenomenon calls for a new paradigm, namely machine unlearning, to make ML models forget about particular data. It turns out that recent works on machine unlearning have not been able to completely solve the problem due to the lack of common frameworks and resources. Therefore, this paper aspires to present a comprehensive examination of machine unlearning's concepts, scenarios, methods, and applications. Specifically, as a category collection of cutting-edge studies, the intention behind this article is to serve as a comprehensive resource for researchers and practitioners seeking an introduction to machine unlearning and its formulations, design criteria, removal requests, algorithms, and applications. In addition, we aim to highlight the key findings, current trends, and new research areas that have not yet featured the use of machine unlearning but could benefit greatly from it. We hope this survey serves as a valuable resource for ML researchers and those seeking to innovate privacy technologies. Our resources are publicly available at https://github.com/tamlhp/awesome-machine-unlearning.
-
Abstract
Data similarity (or distance) computation is a fundamental research topic which fosters a variety of similarity-based machine learning and data mining applications. In big data analytics, it is impractical to compute the exact similarity of data instances due to high computational cost. To this end, the Locality Sensitive Hashing (LSH) technique has been proposed to provide accurate estimators for various similarity measures between sets or vectors in an efficient manner without the learning process. Structured data (e.g., sequences, trees and graphs), which are composed of elements and relations between the elements, are commonly seen in the real world, but the traditional LSH algorithms cannot preserve the structure information represented as relations between elements. In order to conquer the issue, researchers have been devoted to the family of the hierarchical LSH algorithms. In this paper, we explore the present progress of the research into hierarchical LSH from the following perspectives: 1) Data structures, where we review various hierarchical LSH algorithms for three typical data structures and uncover their inherent connections; 2) Applications, where we review the hierarchical LSH algorithms in multiple application scenarios; 3) Challenges, where we discuss some potential challenges as future directions.
-
Abstract
Log-structured merge (LSM) trees offer efficient ingestion by appending incoming data, and thus, are widely used as the storage layer of production NoSQL data stores. To enable competitive read performance, LSM-trees periodically re-organize data to form a tree with levels of exponentially increasing capacity, through iterative compactions. Compactions fundamentally influence the performance of an LSM-engine in terms of write amplification, write throughput, point and range lookup performance, space amplification, and delete performance. Hence, choosing the appropriate compaction strategy is crucial and, at the same time, hard as the LSM-compaction design space is vast, largely unexplored, and has not been formally defined in the literature. As a result, most LSM-based engines use a fixed compaction strategy, typically hand-picked by an engineer, which decides how and when to compact data. In this paper, we present the design space of LSM-compactions, and evaluate state-of-the-art compaction strategies with respect to key performance metrics. Toward this goal, our first contribution is to introduce a set of four design primitives that can formally define any compaction strategy: (i) the compaction trigger, (ii) the data layout, (iii) the compaction granularity, and (iv) the data movement policy. Together, these primitives can synthesize both existing and completely new compaction strategies. Our second contribution is to experimentally analyze 10 compaction strategies. We present 12 observations and 7 high-level takeaway messages, which show how LSM systems can navigate the compaction design space.
-
Abstract
Macroprogramming refers to the theory and practice of conveniently expressing the macro(scopic) behaviour of a system using a single program. Macroprogramming approaches are motivated by the need of effectively capturing global/system-level aspects and the collective behaviour of a set of interacting components, while abstracting over low-level details. In the past, this style of programming has been primarily adopted to describe the data-processing logic in wireless sensor networks; recently, research forums on spatial computing, collective adaptive systems, and Internet-of-Things have provided renewed interest in macro-approaches. However, related contributions are still fragmented and lacking conceptual consistency. Therefore, to foster principled research, an integrated view of the field is provided, together with opportunities and challenges.
-
Abstract
This is Part II of the two-part comprehensive survey devoted to a computing framework most commonly known under the names Hyperdimensional Computing and Vector Symbolic Architectures (HDC/VSA). Both names refer to a family of computational models that use high-dimensional distributed representations and rely on the algebraic properties of their key operations to incorporate the advantages of structured symbolic representations and vector distributed representations. Holographic Reduced Representations is an influential HDC/VSA model that is well-known in the machine learning domain and often used to refer to the whole family. However, for the sake of consistency, we use HDC/VSA to refer to the field. Part I of this survey covered foundational aspects of the field, such as the historical context leading to the development of HDC/VSA, key elements of any HDC/VSA model, known HDC/VSA models, and the transformation of input data of various types into high-dimensional vectors suitable for HDC/VSA. This second part surveys existing applications, the role of HDC/VSA in cognitive computing and architectures, as well as directions for future work. Most of the applications lie within the Machine Learning/Artificial Intelligence domain, however, we also cover other applications to provide a complete picture. The survey is written to be useful for both newcomers and practitioners.
-
Abstract
This two-part comprehensive survey is devoted to a computing framework most commonly known under the names Hyperdimensional Computing and Vector Symbolic Architectures (HDC/VSA). Both names refer to a family of computational models that use high-dimensional distributed representations and rely on the algebraic properties of their key operations to incorporate the advantages of structured symbolic representations and vector distributed representations. Notable models in the HDC/VSA family are Tensor Product Representations, Holographic Reduced Representations, Multiply-Add-Permute, Binary Spatter Codes, and Sparse Binary Distributed Representations but there are other models too. HDC/VSA is a highly interdisciplinary field with connections to computer science, electrical engineering, artificial intelligence, mathematics, and cognitive science. This fact makes it challenging to create a thorough overview of the field. However, due to a surge of new researchers joining the field in recent years, the necessity for a comprehensive survey of the field has become extremely important. Therefore, amongst other aspects of the field, this Part I surveys important aspects such as: known computational models of HDC/VSA and transformations of various input data types to high-dimensional distributed representations. Part II of this survey is devoted to applications, cognitive computing and architectures, as well as directions for future work. The survey is written to be useful for both newcomers and practitioners.
-
Abstract
Learning to act in an environment to maximise rewards is among the brain's key functions. This process has often been conceptualised within the framework of reinforcement learning, which has also gained prominence in machine learning and artificial intelligence (AI) as a way to optimise decision-making. A common aspect of both biological and machine reinforcement learning is the reactivation of previously experienced episodes, referred to as replay. Replay is important for memory consolidation in biological neural networks, and is key to stabilising learning in deep neural networks. Here, we review recent developments concerning the functional roles of replay in the fields of neuroscience and AI. Complementary progress suggests how replay might support learning processes, including generalisation and continual learning, affording opportunities to transfer knowledge across the two fields to advance the understanding of biological and artificial learning and memory.
-
Abstract
Holographic Reduced Representations (HRR) are a method for performing symbolic AI on top of real-valued vectors by associating each vector with an abstract concept, and providing mathematical operations to manipulate vectors as if they were classic symbolic objects. This method has seen little use outside of older symbolic AI work and cognitive science. Our goal is to revisit this approach to understand if it is viable for enabling a hybrid neural-symbolic approach to learning as a differentiable component of a deep learning architecture. HRRs today are not effective in a differentiable solution due to numerical instability, a problem we solve by introducing a projection step that forces the vectors to exist in a well behaved point in space. In doing so we improve the concept retrieval efficacy of HRRs by over $100\times$. Using multi-label classification we demonstrate how to leverage the symbolic HRR properties to develop an output layer and loss function that is able to learn effectively, and allows us to investigate some of the pros and cons of an HRR neuro-symbolic learning approach. Our code can be found at https://github.com/NeuromorphicComputationResearchProgram/Learning-with-Holographic-Reduced-Representations
-
Abstract
Despite recent improvements in open-domain dialogue models, state of the art models are trained and evaluated on short conversations with little context. In contrast, the long-term conversation setting has hardly been studied. In this work we collect and release a human-human dataset consisting of multiple chat sessions whereby the speaking partners learn about each other's interests and discuss the things they have learnt from past sessions. We show how existing models trained on existing datasets perform poorly in this long-term conversation setting in both automatic and human evaluations, and we study long-context models that can perform much better. In particular, we find retrieval-augmented methods and methods with an ability to summarize and recall previous conversations outperform the standard encoder-decoder architectures currently considered state of the art.
-
Abstract
Replay is the reactivation of one or more neural patterns, which are similar to the activation patterns experienced during past waking experiences. Replay was first observed in biological neural networks during sleep, and it is now thought to play a critical role in memory formation, retrieval, and consolidation. Replay-like mechanisms have been incorporated into deep artificial neural networks that learn over time to avoid catastrophic forgetting of previous knowledge. Replay algorithms have been successfully used in a wide range of deep learning methods within supervised, unsupervised, and reinforcement learning paradigms. In this paper, we provide the first comprehensive comparison between replay in the mammalian brain and replay in artificial neural networks. We identify multiple aspects of biological replay that are missing in deep learning systems and hypothesize how they could be utilized to improve artificial neural networks.
-
Abstract
Sybil attacks, in which a large number of adversary-controlled nodes join a network, are a concern for many peer-to-peer database systems, necessitating expensive countermeasures such as proof-of-work. However, there is a category of database applications that are, by design, immune to Sybil attacks because they can tolerate arbitrary numbers of Byzantine-faulty nodes. In this paper, we characterize this category of applications using a consistency model we call Byzantine Eventual Consistency (BEC). We introduce an algorithm that guarantees BEC based on Byzantine causal broadcast, prove its correctness, and demonstrate near-optimal performance in a prototype implementation.
-
Abstract
Contemporary neural networks still fall short of human-level generalization, which extends far beyond our direct experiences. In this paper, we argue that the underlying cause for this shortcoming is their inability to dynamically and flexibly bind information that is distributed throughout the network. This binding problem affects their capacity to acquire a compositional understanding of the world in terms of symbol-like entities (like objects), which is crucial for generalizing in predictable and systematic ways. To address this issue, we propose a unifying framework that revolves around forming meaningful entities from unstructured sensory inputs (segregation), maintaining this separation of information at a representational level (representation), and using these entities to construct new inferences, predictions, and behaviors (composition). Our analysis draws inspiration from a wealth of research in neuroscience and cognitive psychology, and surveys relevant mechanisms from the machine learning literature, to help identify a combination of inductive biases that allow symbolic information processing to emerge naturally in neural networks. We believe that a compositional approach to AI, in terms of grounded symbol-like representations, is of fundamental importance for realizing human-level generalization, and we hope that this paper may contribute towards that goal as a reference and inspiration.
-
Abstract
With data durability, high access speed, low power efficiency and byte addressability, NVMe and SSD, which are acknowledged representatives of emerging storage technologies, have been applied broadly in many areas. However, one key issue with high-performance adoption of these technologies is how to properly define intelligent cache layers such that the performance gap between emerging technologies and main memory can be well bridged. To this end, we propose Phoebe, a reuse-aware reinforcement learning framework for the optimal online caching that is applicable for a wide range of emerging storage models. By continuous interacting with the cache environment and the data stream, Phoebe is capable to extract critical temporal data dependency and relative positional information from a single trace, becoming ever smarter over time. To reduce training overhead during online learning, we utilize periodical training to amortize costs. Phoebe is evaluated on a set of Microsoft cloud storage workloads. Experiment results show that Phoebe is able to close the gap of cache miss rate from LRU and a state-of-the-art online learning based cache policy to the Belady's optimal policy by 70.3% and 52.6%, respectively.
-
Abstract
The Right to be Forgotten is part of the recently enacted General Data Protection Regulation (GDPR) law that affects any data holder that has data on European Union residents. It gives EU residents the ability to request deletion of their personal data, including training records used to train machine learning models. Unfortunately, Deep Neural Network models are vulnerable to information leaking attacks such as model inversion attacks which extract class information from a trained model and membership inference attacks which determine the presence of an example in a model's training data. If a malicious party can mount an attack and learn private information that was meant to be removed, then it implies that the model owner has not properly protected their user's rights and their models may not be compliant with the GDPR law. In this paper, we present two efficient methods that address this question of how a model owner or data holder may delete personal data from models in such a way that they may not be vulnerable to model inversion and membership inference attacks while maintaining model efficacy. We start by presenting a real-world threat model that shows that simply removing training data is insufficient to protect users. We follow that up with two data removal methods, namely Unlearning and Amnesiac Unlearning, that enable model owners to protect themselves against such attacks while being compliant with regulations. We provide extensive empirical analysis that show that these methods are indeed efficient, safe to apply, effectively remove learned information about sensitive data from trained models while maintaining model efficacy.
-
Abstract
Search engines often follow a two-phase paradigm where in the first stage (the retrieval stage) an initial set of documents is retrieved and in the second stage (the re-ranking stage) the documents are re-ranked to obtain the final result list. While deep neural networks were shown to improve the performance of the re-ranking stage in previous works, there is little literature about using deep neural networks to improve the retrieval stage. In this paper, we study the merits of combining deep neural network models and lexical models for the retrieval stage. A hybrid approach, which leverages both semantic (deep neural network-based) and lexical (keyword matching-based) retrieval models, is proposed. We perform an empirical study, using a publicly available TREC collection, which demonstrates the effectiveness of our approach and sheds light on the different characteristics of the semantic approach, the lexical approach, and their combination.
-
Abstract
The goal of text ranking is to generate an ordered list of texts retrieved from a corpus in response to a query. Although the most common formulation of text ranking is search, instances of the task can also be found in many natural language processing applications. This survey provides an overview of text ranking with neural network architectures known as transformers, of which BERT is the best-known example. The combination of transformers and self-supervised pretraining has been responsible for a paradigm shift in natural language processing (NLP), information retrieval (IR), and beyond. In this survey, we provide a synthesis of existing work as a single point of entry for practitioners who wish to gain a better understanding of how to apply transformers to text ranking problems and researchers who wish to pursue work in this area. We cover a wide range of modern techniques, grouped into two high-level categories: transformer models that perform reranking in multi-stage architectures and dense retrieval techniques that perform ranking directly. There are two themes that pervade our survey: techniques for handling long documents, beyond typical sentence-by-sentence processing in NLP, and techniques for addressing the tradeoff between effectiveness (i.e., result quality) and efficiency (e.g., query latency, model and index size). Although transformer architectures and pretraining techniques are recent innovations, many aspects of how they are applied to text ranking are relatively well understood and represent mature techniques. However, there remain many open research questions, and thus in addition to laying out the foundations of pretrained transformers for text ranking, this survey also attempts to prognosticate where the field is heading.
-
Abstract
Existing approaches for open-domain question answering (QA) are typically designed for questions that require either single-hop or multi-hop reasoning, which make strong assumptions of the complexity of questions to be answered. Also, multi-step document retrieval often incurs higher number of relevant but non-supporting documents, which dampens the downstream noise-sensitive reader module for answer extraction. To address these challenges, we propose a unified QA framework to answer any-hop open-domain questions, which iteratively retrieves, reranks and filters documents, and adaptively determines when to stop the retrieval process. To improve the retrieval accuracy, we propose a graph-based reranking model that perform multi-document interaction as the core of our iterative reranking framework. Our method consistently achieves performance comparable to or better than the state-of-the-art on both single-hop and multi-hop open-domain QA datasets, including Natural Questions Open, SQuAD Open, and HotpotQA.
-
Abstract
Symbolic reasoning and neural networks are often considered incompatible approaches. Connectionist models known as Vector Symbolic Architectures (VSAs) can potentially bridge this gap. However, classical VSAs and neural networks are still considered incompatible. VSAs encode symbols by dense pseudo-random vectors, where information is distributed throughout the entire neuron population. Neural networks encode features locally, often forming sparse vectors of neural activation. Following Rachkovskij (2001); Laiho et al. (2015), we explore symbolic reasoning with sparse distributed representations. The core operations in VSAs are dyadic operations between vectors to express variable binding and the representation of sets. Thus, algebraic manipulations enable VSAs to represent and process data structures in a vector space of fixed dimensionality. Using techniques from compressed sensing, we first show that variable binding between dense vectors in VSAs is mathematically equivalent to tensor product binding between sparse vectors, an operation which increases dimensionality. This result implies that dimensionality-preserving binding for general sparse vectors must include a reduction of the tensor matrix into a single sparse vector. Two options for sparsity-preserving variable binding are investigated. One binding method for general sparse vectors extends earlier proposals to reduce the tensor product into a vector, such as circular convolution. The other method is only defined for sparse block-codes, block-wise circular convolution. Our experiments reveal that variable binding for block-codes has ideal properties, whereas binding for general sparse vectors also works, but is lossy, similar to previous proposals. We demonstrate a VSA with sparse block-codes in example applications, cognitive reasoning and classification, and discuss its relevance for neuroscience and neural networks.
-
Abstract
Dense Associative Memories or modern Hopfield networks permit storage and reliable retrieval of an exponentially large (in the dimension of feature space) number of memories. At the same time, their naive implementation is non-biological, since it seemingly requires the existence of many-body synaptic junctions between the neurons. We show that these models are effective descriptions of a more microscopic (written in terms of biological degrees of freedom) theory that has additional (hidden) neurons and only requires two-body interactions between them. For this reason our proposed microscopic theory is a valid model of large associative memory with a degree of biological plausibility. The dynamics of our network and its reduced dimensional equivalent both minimize energy (Lyapunov) functions. When certain dynamical variables (hidden neurons) are integrated out from our microscopic theory, one can recover many of the models that were previously discussed in the literature, e.g. the model presented in "Hopfield Networks is All You Need" paper. We also provide an alternative derivation of the energy function and the update rule proposed in the aforementioned paper and clarify the relationships between various models of this class.
-
Abstract
Caching techniques are widely used in the era of cloud computing from applications, such as Web caches to infrastructures, Memcached and memory caches in computer architectures. Prediction of cached data can greatly help improve cache management and performance. The recent advancement of deep learning techniques enables the design of novel intelligent cache replacement policies. In this work, we propose a learning-aided approach to predict future data accesses. We find that a powerful LSTM-based recurrent neural network model can provide high prediction accuracy based on only a cache trace as input. The high accuracy results from a carefully crafted locality-driven feature design. Inspired by the high prediction accuracy, we propose a pseudo OPT policy and evaluate it upon 13 real-world storage workloads from Microsoft Research. Results demonstrate that the new cache policy improves state-of-art practical policies by up to 19.2% and incurs only 2.3% higher miss ratio than OPT on average.
-
Abstract
Novelty detection, i.e., identifying whether a given sample is drawn from outside the training distribution, is essential for reliable machine learning. To this end, there have been many attempts at learning a representation well-suited for novelty detection and designing a score based on such representation. In this paper, we propose a simple, yet effective method named contrasting shifted instances (CSI), inspired by the recent success on contrastive learning of visual representations. Specifically, in addition to contrasting a given sample with other instances as in conventional contrastive learning methods, our training scheme contrasts the sample with distributionally-shifted augmentations of itself. Based on this, we propose a new detection score that is specific to the proposed training scheme. Our experiments demonstrate the superiority of our method under various novelty detection scenarios, including unlabeled one-class, unlabeled multi-class and labeled multi-class settings, with various image benchmark datasets. Code and pre-trained models are available at https://github.com/alinlab/CSI.
-
Abstract
We introduce a modern Hopfield network with continuous states and a corresponding update rule. The new Hopfield network can store exponentially (with the dimension of the associative space) many patterns, retrieves the pattern with one update, and has exponentially small retrieval errors. It has three types of energy minima (fixed points of the update): (1) global fixed point averaging over all patterns, (2) metastable states averaging over a subset of patterns, and (3) fixed points which store a single pattern. The new update rule is equivalent to the attention mechanism used in transformers. This equivalence enables a characterization of the heads of transformer models. These heads perform in the first layers preferably global averaging and in higher layers partial averaging via metastable states. The new modern Hopfield network can be integrated into deep learning architectures as layers to allow the storage of and access to raw input data, intermediate results, or learned prototypes. These Hopfield layers enable new ways of deep learning, beyond fully-connected, convolutional, or recurrent networks, and provide pooling, memory, association, and attention mechanisms. We demonstrate the broad applicability of the Hopfield layers across various domains. Hopfield layers improved state-of-the-art on three out of four considered multiple instance learning problems as well as on immune repertoire classification with several hundreds of thousands of instances. On the UCI benchmark collections of small classification tasks, where deep learning methods typically struggle, Hopfield layers yielded a new state-of-the-art when compared to different machine learning methods. Finally, Hopfield layers achieved state-of-the-art on two drug design datasets. The implementation is available at: https://github.com/ml-jku/hopfield-layers
-
Abstract
Data-intensive applications fueled the evolution of log structured merge (LSM) based key-value engines that employ the out-of-place paradigm to support high ingestion rates with low read/write interference. These benefits, however, come at the cost of treating deletes as a second-class citizen. A delete inserts a tombstone that invalidates older instances of the deleted key. State-of-the-art LSM engines do not provide guarantees as to how fast a tombstone will propagate to persist the deletion. Further, LSM engines only support deletion on the sort key. To delete on another attribute (e.g., timestamp), the entire tree is read and re-written. We highlight that fast persistent deletion without affecting read performance is key to support: (i) streaming systems operating on a window of data, (ii) privacy with latency guarantees on the right-to-be-forgotten, and (iii) en masse cloud deployment of data systems that makes storage a precious resource. To address these challenges, in this paper, we build a new key-value storage engine, Lethe, that uses a very small amount of additional metadata, a set of new delete-aware compaction policies, and a new physical data layout that weaves the sort and the delete key order. We show that Lethe supports any user-defined threshold for the delete persistence latency offering higher read throughput ($1.17-1.4\times$) and lower space amplification ($2.1-9.8\times$), with a modest increase in write amplification (between $4\%$ and $25\%$). In addition, Lethe supports efficient range deletes on a secondary delete key by dropping entire data pages without sacrificing read performance nor employing a costly full tree merge.
-
Abstract
Coresets are small data summaries that are sufficient for model training. They can be maintained online, enabling efficient handling of large data streams under resource constraints. However, existing constructions are limited to simple models such as k-means and logistic regression. In this work, we propose a novel coreset construction via cardinality-constrained bilevel optimization. We show how our framework can efficiently generate coresets for deep neural networks, and demonstrate its empirical benefits in continual learning and in streaming settings.
-
Abstract
Evidence-based decision-making entails collecting (costly) observations about an underlying phenomenon of interest, and subsequently committing to an (informed) decision on the basis of accumulated evidence. In this setting, active sensing is the goal-oriented problem of efficiently selecting which acquisitions to make, and when and what decision to settle on. As its complement, inverse active sensing seeks to uncover an agent's preferences and strategy given their observable decision-making behavior. In this paper, we develop an expressive, unified framework for the general setting of evidence-based decision-making under endogenous, context-dependent time pressure---which requires negotiating (subjective) tradeoffs between accuracy, speediness, and cost of information. Using this language, we demonstrate how it enables modeling intuitive notions of surprise, suspense, and optimality in decision strategies (the forward problem). Finally, we illustrate how this formulation enables understanding decision-making behavior by quantifying preferences implicit in observed decision strategies (the inverse problem).
-
Abstract
Replay in neural networks involves training on sequential data with memorized samples, which counteracts forgetting of previous behavior caused by non-stationarity. We present a method where these auxiliary samples are generated on the fly, given only the model that is being trained for the assessed objective, without extraneous buffers or generator networks. Instead the implicit memory of learned samples within the assessed model itself is exploited. Furthermore, whereas existing work focuses on reinforcing the full seen data distribution, we show that optimizing for not forgetting calls for the generation of samples that are specialized to each real training batch, which is more efficient and scalable. We consider high-level parallels with the brain, notably the use of a single model for inference and recall, the dependency of recalled samples on the current environment batch, top-down modulation of activations and learning, abstract recall, and the dependency between the degree to which a task is learned and the degree to which it is recalled. These characteristics emerge naturally from the method without being controlled for.
-
Abstract
Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still limited, and hence on knowledge-intensive tasks, their performance lags behind task-specific architectures. Additionally, providing provenance for their decisions and updating their world knowledge remain open research problems. Pre-trained models with a differentiable access mechanism to explicit non-parametric memory can overcome this issue, but have so far been only investigated for extractive downstream tasks. We explore a general-purpose fine-tuning recipe for retrieval-augmented generation (RAG) -- models which combine pre-trained parametric and non-parametric memory for language generation. We introduce RAG models where the parametric memory is a pre-trained seq2seq model and the non-parametric memory is a dense vector index of Wikipedia, accessed with a pre-trained neural retriever. We compare two RAG formulations, one which conditions on the same retrieved passages across the whole generated sequence, the other can use different passages per token. We fine-tune and evaluate our models on a wide range of knowledge-intensive NLP tasks and set the state-of-the-art on three open domain QA tasks, outperforming parametric seq2seq models and task-specific retrieve-and-extract architectures. For language generation tasks, we find that RAG models generate more specific, diverse and factual language than a state-of-the-art parametric-only seq2seq baseline.
-
Abstract
Various machine learning tasks can benefit from access to external information of different modalities, such as text and images. Recent work has focused on learning architectures with large memories capable of storing this knowledge. We propose augmenting generative Transformer neural networks with KNN-based Information Fetching (KIF) modules. Each KIF module learns a read operation to access fixed external knowledge. We apply these modules to generative dialog modeling, a challenging task where information must be flexibly retrieved and incorporated to maintain the topic and flow of conversation. We demonstrate the effectiveness of our approach by identifying relevant knowledge required for knowledgeable but engaging dialog from Wikipedia, images, and human-written dialog utterances, and show that leveraging this retrieved information improves model performance, measured by automatic and human evaluation.
-
Abstract
In this paper, we propose a multi-timescale replay (MTR) buffer for improving continual learning in RL agents faced with environments that are changing continuously over time at timescales that are unknown to the agent. The basic MTR buffer comprises a cascade of sub-buffers that accumulate experiences at different timescales, enabling the agent to improve the trade-off between adaptation to new data and retention of old knowledge. We also combine the MTR framework with invariant risk minimization, with the idea of encouraging the agent to learn a policy that is robust across the various environments it encounters over time. The MTR methods are evaluated in three different continual learning settings on two continuous control tasks and, in many cases, show improvement over the baselines.
-
Abstract
Recent progress in Natural Language Understanding (NLU) is driving fast-paced advances in Information Retrieval (IR), largely owed to fine-tuning deep language models (LMs) for document ranking. While remarkably effective, the ranking models based on these LMs increase computational cost by orders of magnitude over prior approaches, particularly as they must feed each query-document pair through a massive neural network to compute a single relevance score. To tackle this, we present ColBERT, a novel ranking model that adapts deep LMs (in particular, BERT) for efficient retrieval. ColBERT introduces a late interaction architecture that independently encodes the query and the document using BERT and then employs a cheap yet powerful interaction step that models their fine-grained similarity. By delaying and yet retaining this fine-granular interaction, ColBERT can leverage the expressiveness of deep LMs while simultaneously gaining the ability to pre-compute document representations offline, considerably speeding up query processing. Beyond reducing the cost of re-ranking the documents retrieved by a traditional model, ColBERT's pruning-friendly interaction mechanism enables leveraging vector-similarity indexes for end-to-end retrieval directly from a large document collection. We extensively evaluate ColBERT using two recent passage search datasets. Results show that ColBERT's effectiveness is competitive with existing BERT-based models (and outperforms every non-BERT baseline), while executing two orders-of-magnitude faster and requiring four orders-of-magnitude fewer FLOPs per query.
-
Abstract
Anomaly detection for time-series data has been an important research field for a long time. Seminal work on anomaly detection methods has been focussing on statistical approaches. In recent years an increasing number of machine learning algorithms have been developed to detect anomalies on time-series. Subsequently, researchers tried to improve these techniques using (deep) neural networks. In the light of the increasing number of anomaly detection methods, the body of research lacks a broad comparative evaluation of statistical, machine learning and deep learning methods. This paper studies 20 univariate anomaly detection methods from the all three categories. The evaluation is conducted on publicly available datasets, which serve as benchmarks for time-series anomaly detection. By analyzing the accuracy of each method as well as the computation time of the algorithms, we provide a thorough insight about the performance of these anomaly detection approaches, alongside some general notion of which method is suited for a certain type of data.
-
Abstract
Over the past years, there has been an increasing number of key-value (KV) store designs, each optimizing for a different set of requirements. Furthermore, with the advancements of storage technology the design space of KV stores has become even more complex. More recent KV-store designs target fast storage devices, such as SSDs and NVM. Most of these designs aim to reduce amplification during data reorganization by taking advantage of device characteristics. However, until today most analysis of KV-store designs is experimental and limited to specific design points. This makes it difficult to compare tradeoffs across different designs, find optimal configurations and guide future KV-store design. In this paper, we introduce the Variable Amplification- Throughput analysis (VAT) to calculate insert-path amplification and its impact on multi-level KV-store performance.We use VAT to express the behavior of several existing design points and to explore tradeoffs that are not possible or easy to measure experimentally. VAT indicates that by inserting randomness in the insert-path, KV stores can reduce amplification by more than 10x for fast storage devices. Techniques, such as key-value separation and tiering compaction, reduce amplification by 10x and 5x, respectively. Additionally, VAT predicts that the advancements in device technology towards NVM, reduces the benefits from both using key-value separation and tiering.
-
Abstract
Term frequency-inverse document frequency, or TF-IDF for short, and its many variants form a class of term weighting functions the members of which are widely used in text analysis applications. While TF-IDF was originally proposed as a heuristic, theoretical justifications grounded in information theory, probability, and the divergence from randomness paradigm have been advanced. In this work, we present an empirical study showing that TF-IDF corresponds very nearly with the hypergeometric test of statistical significance on selected real-data document retrieval, summarization, and classification tasks. These findings suggest that a fundamental mathematical connection between TF-IDF and the negative logarithm of the hypergeometric test P-value (i.e., a hypergeometric distribution tail probability) remains to be elucidated. We advance the empirical analyses herein as a first step toward explaining the long-standing effectiveness of TF-IDF from a statistical significance testing lens. It is our aspiration that these results will open the door to the systematic evaluation of significance testing derived term weighting functions in text analysis applications.
-
Abstract
Artificial neural networks are well-known to be susceptible to catastrophic forgetting when continually learning from sequences of tasks. Various continual (or "incremental") learning approaches have been proposed to avoid catastrophic forgetting, but they are typically adversary agnostic, i.e., they do not consider the possibility of a malicious attack. In this effort, we explore the vulnerability of Elastic Weight Consolidation (EWC), a popular continual learning algorithm for avoiding catastrophic forgetting. We show that an intelligent adversary can bypass the EWC's defenses, and instead cause gradual and deliberate forgetting by introducing small amounts of misinformation to the model during training. We demonstrate such an adversary's ability to assume control of the model via injection of "backdoor" attack samples on both permuted and split benchmark variants of the MNIST dataset. Importantly, once the model has learned the adversarial misinformation, the adversary can then control the amount of forgetting of any task. Equivalently, the malicious actor can create a "false memory" about any task by inserting carefully-designed backdoor samples to any fraction of the test instances of that task. Perhaps most damaging, we show this vulnerability to be very acute; neural network memory can be easily compromised with the addition of backdoor samples into as little as 1% of the training data of even a single task.
-
Abstract
Language model pre-training has been shown to capture a surprising amount of world knowledge, crucial for NLP tasks such as question answering. However, this knowledge is stored implicitly in the parameters of a neural network, requiring ever-larger networks to cover more facts. To capture knowledge in a more modular and interpretable way, we augment language model pre-training with a latent knowledge retriever, which allows the model to retrieve and attend over documents from a large corpus such as Wikipedia, used during pre-training, fine-tuning and inference. For the first time, we show how to pre-train such a knowledge retriever in an unsupervised manner, using masked language modeling as the learning signal and backpropagating through a retrieval step that considers millions of documents. We demonstrate the effectiveness of Retrieval-Augmented Language Model pre-training (REALM) by fine-tuning on the challenging task of Open-domain Question Answering (Open-QA). We compare against state-of-the-art models for both explicit and implicit knowledge storage on three popular Open-QA benchmarks, and find that we outperform all previous methods by a significant margin (4-16% absolute accuracy), while also providing qualitative benefits such as interpretability and modularity.
-
Abstract
Classical approaches for one-class problems such as one-class SVM and isolation forest require careful feature engineering when applied to structured domains like images. State-of-the-art methods aim to leverage deep learning to learn appropriate features via two main approaches. The first approach based on predicting transformations (Golan & El-Yaniv, 2018; Hendrycks et al., 2019a) while successful in some domains, crucially depends on an appropriate domain-specific set of transformations that are hard to obtain in general. The second approach of minimizing a classical one-class loss on the learned final layer representations, e.g., DeepSVDD (Ruff et al., 2018) suffers from the fundamental drawback of representation collapse. In this work, we propose Deep Robust One-Class Classification (DROCC) that is both applicable to most standard domains without requiring any side-information and robust to representation collapse. DROCC is based on the assumption that the points from the class of interest lie on a well-sampled, locally linear low dimensional manifold. Empirical evaluation demonstrates that DROCC is highly effective in two different one-class problem settings and on a range of real-world datasets across different domains: tabular data, images (CIFAR and ImageNet), audio, and time-series, offering up to 20% increase in accuracy over the state-of-the-art in anomaly detection. Code is available at https://github.com/microsoft/EdgeML.
-
Abstract
Nearest neighbors is a successful and long-standing technique for anomaly detection. Significant progress has been recently achieved by self-supervised deep methods (e.g. RotNet). Self-supervised features however typically under-perform Imagenet pre-trained features. In this work, we investigate whether the recent progress can indeed outperform nearest-neighbor methods operating on an Imagenet pretrained feature space. The simple nearest-neighbor based-approach is experimentally shown to outperform self-supervised methods in: accuracy, few shot generalization, training time and noise robustness while making fewer assumptions on image distributions.
-
Abstract
With the rapid evolution of wireless mobile devices, there emerges an increased need to design effective collaboration mechanisms between intelligent agents, so as to gradually approach the final collective objective through continuously learning from the environment based on their individual observations. In this regard, independent reinforcement learning (IRL) is often deployed in multi-agent collaboration to alleviate the problem of a non-stationary learning environment. However, behavioral strategies of intelligent agents in IRL can only be formulated upon their local individual observations of the global environment, and appropriate communication mechanisms must be introduced to reduce their behavioral localities. In this paper, we address the problem of communication between intelligent agents in IRL by jointly adopting mechanisms with two different scales. For the large scale, we introduce the stigmergy mechanism as an indirect communication bridge between independent learning agents, and carefully design a mathematical method to indicate the impact of digital pheromone. For the small scale, we propose a conflict-avoidance mechanism between adjacent agents by implementing an additionally embedded neural network to provide more opportunities for participants with higher action priorities. In addition, we present a federal training method to effectively optimize the neural network of each agent in a decentralized manner. Finally, we establish a simulation scenario in which a number of mobile agents in a certain area move automatically to form a specified target shape. Extensive simulations demonstrate the effectiveness of our proposed method.
-
Abstract
People learn throughout life. However, incrementally updating conventional neural networks leads to catastrophic forgetting. A common remedy is replay, which is inspired by how the brain consolidates memory. Replay involves fine-tuning a network on a mixture of new and old instances. While there is neuroscientific evidence that the brain replays compressed memories, existing methods for convolutional networks replay raw images. Here, we propose REMIND, a brain-inspired approach that enables efficient replay with compressed representations. REMIND is trained in an online manner, meaning it learns one example at a time, which is closer to how humans learn. Under the same constraints, REMIND outperforms other methods for incremental class learning on the ImageNet ILSVRC-2012 dataset. We probe REMIND's robustness to data ordering schemes known to induce catastrophic forgetting. We demonstrate REMIND's generality by pioneering online learning for Visual Question Answering (VQA).
-
Abstract
Field-based coordination has been proposed as a model for coordinating collective adaptive systems, promoting a view of distributed computations as functions manipulating data structures spread over space and evolving over time, called computational fields. The field calculus is a formal foundation for field computations, providing specific constructs for evolution (time) and neighbor interaction (space), which are handled by separate operators (called rep and nbr, respectively). This approach, however, intrinsically limits the speed of information propagation that can be achieved by their combined use. In this paper, we propose a new field-based coordination operator called share, which captures the space-time nature of field computations in a single operator that declaratively achieves: (i) observation of neighbors' values; (ii) reduction to a single local value; and (iii) update and converse sharing to neighbors of a local variable. We show that for an important class of self-stabilising computations, share can replace all occurrences of rep and nbr constructs. In addition to conceptual economy, use of the share operator also allows many prior field calculus algorithms to be greatly accelerated, which we validate empirically with simulations of frequently used network propagation and collection algorithms.
-
Abstract
This study investigates the use of reinforcement learning to guide a general purpose cache manager decisions. Cache managers directly impact the overall performance of computer systems. They govern decisions about which objects should be cached, the duration they should be cached for, and decides on which objects to evict from the cache if it is full. These three decisions impact both the cache hit rate and size of the storage that is needed to achieve that cache hit rate. An optimal cache manager will avoid unnecessary operations, maximise the cache hit rate which results in fewer round trips to a slower backend storage system, and minimise the size of storage needed to achieve a high hit-rate. This project investigates using reinforcement learning in cache management by designing three separate agents for each of the cache manager tasks. Furthermore, the project investigates two advanced reinforcement learning architectures for multi-decision problems: a single multi-task agent and a multi-agent. We also introduce a framework to simplify the modelling of computer systems problems as a reinforcement learning task. The framework abstracts delayed experiences observations and reward assignment in computer systems while providing a flexible way to scale to multiple agents. Simulation results based on an established database benchmark system show that reinforcement learning agents can achieve a higher cache hit rate over heuristic driven algorithms while minimising the needed space. They are also able to adapt to a changing workload and dynamically adjust their caching strategy accordingly. The proposed cache manager model is generic and applicable to other types of caches, such as file system caches. This project is the first, to our knowledge, to model cache manager decisions as a multi-task control problem.
-
Abstract
The reinforcement learning (RL) research area is very active, with an important number of new contributions; especially considering the emergent field of deep RL (DRL). However a number of scientific and technical challenges still need to be addressed, amongst which we can mention the ability to abstract actions or the difficulty to explore the environment which can be addressed by intrinsic motivation (IM). In this article, we provide a survey on the role of intrinsic motivation in DRL. We categorize the different kinds of intrinsic motivations and detail for each category, its advantages and limitations with respect to the mentioned challenges. Additionnally, we conduct an in-depth investigation of substantial current research questions, that are currently under study or not addressed at all in the considered research area of DRL. We choose to survey these research works, from the perspective of learning how to achieve tasks. We suggest then, that solving current challenges could lead to a larger developmental architecture which may tackle most of the tasks. We describe this developmental architecture on the basis of several building blocks composed of a RL algorithm and an IM module compressing information.
-
Abstract
We introduce the concept of design continuums for the data layout of key-value stores. A design continuum unifies major distinct data structure designs under the same model. The critical insight and potential long-term impact is that such unifying models 1) render what we consider up to now as fundamentally different data structures to be seen as views of the very same overall design space, and 2) allow seeing new data structure designs with performance properties that are not feasible by existing designs. The core intuition behind the construction of design continuums is that all data structures arise from the very same set of fundamental design principles, i.e., a small set of data layout design concepts out of which we can synthesize any design that exists in the literature as well as new ones. We show how to construct, evaluate, and expand, design continuums and we also present the first continuum that unifies major data structure designs, i.e., B+tree, B-epsilon-tree, LSM-tree, and LSH-table. The practical benefit of a design continuum is that it creates a fast inference engine for the design of data structures. For example, we can predict near instantly how a specific design change in the underlying storage of a data system would affect performance, or reversely what would be the optimal data structure (from a given set of designs) given workload characteristics and a memory budget. In turn, these properties allow us to envision a new class of self-designing key-value stores with a substantially improved ability to adapt to workload and hardware changes by transitioning between drastically different data structure designs to assume a diverse set of performance properties at will.
-
Abstract
Mutual information has been successfully adopted in filter feature-selection methods to assess both the relevancy of a subset of features in predicting the target variable and the redundancy with respect to other variables. However, existing algorithms are mostly heuristic and do not offer any guarantee on the proposed solution. In this paper, we provide novel theoretical results showing that conditional mutual information naturally arises when bounding the ideal regression/classification errors achieved by different subsets of features. Leveraging on these insights, we propose a novel stopping condition for backward and forward greedy methods which ensures that the ideal prediction error using the selected feature subset remains bounded by a user-specified threshold. We provide numerical simulations to support our theoretical claims and compare to common heuristic methods.
-
Abstract
Intense recent discussions have focused on how to provide individuals with control over when their data can and cannot be used --- the EU's Right To Be Forgotten regulation is an example of this effort. In this paper we initiate a framework studying what to do when it is no longer permissible to deploy models derivative from specific user data. In particular, we formulate the problem of efficiently deleting individual data points from trained machine learning models. For many standard ML models, the only way to completely remove an individual's data is to retrain the whole model from scratch on the remaining data, which is often not computationally practical. We investigate algorithmic principles that enable efficient data deletion in ML. For the specific setting of k-means clustering, we propose two provably efficient deletion algorithms which achieve an average of over 100X improvement in deletion efficiency across 6 datasets, while producing clusters of comparable statistical quality to a canonical k-means++ baseline.
-
Abstract
Hyperdimensional computing (HDC) is an emerging computational framework that takes inspiration from attributes of neuronal circuits such as hyperdimensionality, fully distributed holographic representation, and (pseudo)randomness. When employed for machine learning tasks such as learning and classification, HDC involves manipulation and comparison of large patterns within memory. Moreover, a key attribute of HDC is its robustness to the imperfections associated with the computational substrates on which it is implemented. It is therefore particularly amenable to emerging non-von Neumann paradigms such as in-memory computing, where the physical attributes of nanoscale memristive devices are exploited to perform computation in place. Here, we present a complete in-memory HDC system that achieves a near optimum trade-off between design complexity and classification accuracy based on three prototypical HDC related learning tasks, namely, language classification, news classification, and hand gesture recognition from electromyography signals. Comparable accuracies to software implementations are demonstrated, experimentally, using 760,000 phase-change memory devices performing analog in-memory computing.
-
Abstract
Data selection methods, such as active learning and core-set selection, are useful tools for machine learning on large datasets. However, they can be prohibitively expensive to apply in deep learning because they depend on feature representations that need to be learned. In this work, we show that we can greatly improve the computational efficiency by using a small proxy model to perform data selection (e.g., selecting data points to label for active learning). By removing hidden layers from the target model, using smaller architectures, and training for fewer epochs, we create proxies that are an order of magnitude faster to train. Although these small proxy models have higher error rates, we find that they empirically provide useful signals for data selection. We evaluate this "selection via proxy" (SVP) approach on several data selection tasks across five datasets: CIFAR10, CIFAR100, ImageNet, Amazon Review Polarity, and Amazon Review Full. For active learning, applying SVP can give an order of magnitude improvement in data selection runtime (i.e., the time it takes to repeatedly train and select points) without significantly increasing the final error (often within 0.1%). For core-set selection on CIFAR10, proxies that are over 10x faster to train than their larger, more accurate targets can remove up to 50% of the data without harming the final accuracy of the target, leading to a 1.6x end-to-end training time improvement.
-
Abstract
The Log-Structured Merge-Tree (LSM-tree) has been widely adopted for use in modern NoSQL systems for its superior write performance. Despite the popularity of LSM-trees, they have been criticized for suffering from write stalls and large performance variances due to the inherent mismatch between their fast in-memory writes and slow background I/O operations. In this paper, we use a simple yet effective two-phase experimental approach to evaluate write stalls for various LSM-tree designs. We further explore the design choices of LSM merge schedulers to minimize write stalls given an I/O bandwidth budget. We have conducted extensive experiments in the context of the Apache AsterixDB system and we present the results here.
-
Abstract
We present apricot, an open source Python package for selecting representative subsets from large data sets using submodular optimization. The package implements an efficient greedy selection algorithm that offers strong theoretical guarantees on the quality of the selected set. Two submodular set functions are implemented in apricot: facility location, which is broadly applicable but requires memory quadratic in the number of examples in the data set, and a feature-based function that is less broadly applicable but can scale to millions of examples. Apricot is extremely efficient, using both algorithmic speedups such as the lazy greedy algorithm and code optimizers such as numba. We demonstrate the use of subset selection by training machine learning models to comparable accuracy using either the full data set or a representative subset thereof. This paper presents an explanation of submodular selection, an overview of the features in apricot, and an application to several data sets. The code and tutorial Jupyter notebooks are available at https://github.com/jmschrei/apricot
-
Abstract
The most important obstacles facing multi-document summarization include excessive redundancy in source descriptions and the looming shortage of training data. These obstacles prevent encoder-decoder models from being used directly, but optimization-based methods such as determinantal point processes (DPPs) are known to handle them well. In this paper we seek to strengthen a DPP-based method for extractive multi-document summarization by presenting a novel similarity measure inspired by capsule networks. The approach measures redundancy between a pair of sentences based on surface form and semantic information. We show that our DPP system with improved similarity measure performs competitively, outperforming strong summarization baselines on benchmark datasets. Our findings are particularly meaningful for summarizing documents created by multiple authors containing redundant yet lexically diverse expressions.
-
Abstract
Distributed storage systems employ replication to improve performance and reliability. To provide low latency data access, replicas are often required to accept updates without coordination with each other, and the updates are then propagated asynchronously. This brings the critical challenge of conflict resolution among concurrent updates. Conflict-free Replicated Data Type (CRDT) is a principled approach to addressing this challenge. However, existing CRDT designs are tricky, and hard to be generalized to other data types. A design framework is in great need to guide the systematic design of new CRDTs. To address this challenge, we propose RWF -- the Remove-Win design Framework for CRDTs. RWF leverages the simple but powerful remove-win strategy to resolve conflicting updates, and provides generic design for a variety of data container types. Two exemplar implementations following RWF are given over the Redis data type store, which demonstrate the effectiveness of RWF. Performance measurements of our implementations further show the efficiency of CRDT designs following RWF.
-
Abstract
General solutions of state machine replication have to ensure that all replicas apply the same commands in the same order, even in the presence of failures. Such strict ordering incurs high synchronization costs caused by distributed consensus or by the use of a leader. This paper presents a protocol for linearizable state machine replication of conflict-free replicated data types (CRDTs) that neither requires consensus nor a leader. By leveraging the properties of state-based CRDTs - in particular, the monotonic growth of a join semilattice - synchronization overhead is greatly reduced. As a result, updates only need a single round trip and modify the state 'in-place' without the need for a log. Furthermore, the message size overhead for coordination consists of a single counter per message. For queries, we guarantee finite writes termination. We show in an experimental evaluation that more than 99 % of queries can be handled in one to three round trips under highly concurrent accesses. Our protocol achieves high throughput without auxiliary processes such as command log management or leader election. Thus, it is well suited for practical scenarios that need linearizable access to CRDT data on a fine-granular scale.
-
Abstract
OT (Operational Transformation) was invented for supporting real-time co-editors in the late 1980s and has evolved to become a core technique used in today's working co-editors and adopted in major industrial products. CRDT (Commutative Replicated Data Type) for co-editors was first proposed around 2006, under the name of WOOT (WithOut Operational Transformation). Follow-up CRDT variations are commonly labeled as "post-OT" techniques and have made broad claims of superiority over OT solutions, in terms of correctness, time and space complexity, simplicity, etc. Over one decade later, however, OT remains the choice for building the vast majority of co-editors, whereas CRDT is rarely found in working co-editors. Why? To seek truth from facts, we set out to conduct a comprehensive and critical review of representative OT and CRDT solutions and working co-editors based on them. From this work, we have made important discoveries about OT and CRDT, and revealed facts and evidences that refute CRDT claims over OT on all accounts. We present our discoveries in three related and complementary articles. In prior two articles, we have revealed the similarities of OT and CRDT in following the same general transformation approach in co-editors, and their real differences in correctness and complexity. In this article, we examine the role of building working co-editors in shaping OT and CRDT research and solutions, and consequential differences in the choice between OT and CRDT in real world co-editors and industry products. In particular, we review the evolution of co-editors from research vehicles to real world applications, and discuss representative OT-based co-editors and alternative approaches in industry products and open source projects. Moreover, we evaluate CRDT-based co-editors in relation to published CRDT solutions, and clarify some myths surrounding "peer-to-peer" co-editing.
-
Abstract
We look at one important category of distributed applications characterized by the existence of multiple collaborating, and competing, components sharing mutable, long-lived, replicated objects. The problem addressed by our work is that of object state synchronization among the components. As an organizing principle for replicated objects, we formally specify the Global Object Tracker (GoT) model, an object-oriented programming model based on causal consistency with application-level conflict resolution strategies, whose elements and interfaces mirror those found in decentralized version control systems: a version graph, working data, diffs, commit, checkout, fetch, push, and merge. We have implemented GoT in a framework called Spacetime, written in Python. In its purest form, GoT is impractical for real systems, because of the unbounded growth of the version graph and because passing diff'ed histories over the network makes remote communication too slow. We present our solution to these problems that adds some constraints to GoT applications, but that makes the model feasible in practice. We present a performance analysis of Spacetime for representative workloads, which shows that the additional constraints added to GoT make it not just feasible, but viable for real applications.
-
Abstract
Despite huge success, deep networks are unable to learn effectively in sequential multitask learning settings as they forget the past learned tasks after learning new tasks. Inspired from complementary learning systems theory, we address this challenge by learning a generative model that couples the current task to the past learned tasks through a discriminative embedding space. We learn an abstract level generative distribution in the embedding that allows the generation of data points to represent the experience. We sample from this distribution and utilize experience replay to avoid forgetting and simultaneously accumulate new knowledge to the abstract distribution in order to couple the current task with past experience. We demonstrate theoretically and empirically that our framework learns a distribution in the embedding that is shared across all task and as a result tackles catastrophic forgetting.
-
Abstract
The utility of learning a dynamics/world model of the environment in reinforcement learning has been shown in a many ways. When using neural networks, however, these models suffer catastrophic forgetting when learned in a lifelong or continual fashion. Current solutions to the continual learning problem require experience to be segmented and labeled as discrete tasks, however, in continuous experience it is generally unclear what a sufficient segmentation of tasks would be. Here we propose a method to continually learn these internal world models through the interleaving of internally generated episodes of past experiences (i.e., pseudo-rehearsal). We show this method can sequentially learn unsupervised temporal prediction, without task labels, in a disparate set of Atari games. Empirically, this interleaving of the internally generated rollouts with the external environment's observations leads to a consistent reduction in temporal prediction loss compared to non-interleaved learning and is preserved over repeated random exposures to various tasks. Similarly, using a network distillation approach, we show that modern policy gradient based reinforcement learning algorithms can use this internal model to continually learn to optimize reward based on the world model's representation of the environment.
-
Abstract
In many machine learning applications, one needs to interactively select a sequence of items (e.g., recommending movies based on a user's feedback) or make sequential decisions in a certain order (e.g., guiding an agent through a series of states). Not only do sequences already pose a dauntingly large search space, but we must also take into account past observations, as well as the uncertainty of future outcomes. Without further structure, finding an optimal sequence is notoriously challenging, if not completely intractable. In this paper, we view the problem of adaptive and sequential decision making through the lens of submodularity and propose an adaptive greedy policy with strong theoretical guarantees. Additionally, to demonstrate the practical utility of our results, we run experiments on Amazon product recommendation and Wikipedia link prediction tasks.
-
Abstract
We study the problem of reducing the amount of labeled training data required to train supervised classification models. We approach it by leveraging Active Learning, through sequential selection of examples which benefit the model most. Selecting examples one by one is not practical for the amount of training examples required by the modern Deep Learning models. We consider the mini-batch Active Learning setting, where several examples are selected at once. We present an approach which takes into account both informativeness of the examples for the model, as well as the diversity of the examples in a mini-batch. By using the well studied K-means clustering algorithm, this approach scales better than the previously proposed approaches, and achieves comparable or better performance.
-
Abstract
A grand challenge in reinforcement learning is intelligent exploration, especially when rewards are sparse or deceptive. Two Atari games serve as benchmarks for such hard-exploration domains: Montezuma's Revenge and Pitfall. On both games, current RL algorithms perform poorly, even those with intrinsic motivation, which is the dominant method to improve performance on hard-exploration domains. To address this shortfall, we introduce a new algorithm called Go-Explore. It exploits the following principles: (1) remember previously visited states, (2) first return to a promising state (without exploration), then explore from it, and (3) solve simulated environments through any available means (including by introducing determinism), then robustify via imitation learning. The combined effect of these principles is a dramatic performance improvement on hard-exploration problems. On Montezuma's Revenge, Go-Explore scores a mean of over 43k points, almost 4 times the previous state of the art. Go-Explore can also harness human-provided domain knowledge and, when augmented with it, scores a mean of over 650k points on Montezuma's Revenge. Its max performance of nearly 18 million surpasses the human world record, meeting even the strictest definition of "superhuman" performance. On Pitfall, Go-Explore with domain knowledge is the first algorithm to score above zero. Its mean score of almost 60k points exceeds expert human performance. Because Go-Explore produces high-performing demonstrations automatically and cheaply, it also outperforms imitation learning work where humans provide solution demonstrations. Go-Explore opens up many new research directions into improving it and weaving its insights into current RL algorithms. It may also enable progress on previously unsolvable hard-exploration problems in many domains, especially those that harness a simulator during training (e.g. robotics).
-
Abstract
Recently, the Log-Structured Merge-tree (LSM-tree) has been widely adopted for use in the storage layer of modern NoSQL systems. Because of this, there have been a large number of research efforts, from both the database community and the operating systems community, that try to improve various aspects of LSM-trees. In this paper, we provide a survey of recent research efforts on LSM-trees so that readers can learn the state-of-the-art in LSM-based storage techniques. We provide a general taxonomy to classify the literature of LSM-trees, survey the efforts in detail, and discuss their strengths and trade-offs. We further survey several representative LSM-based open-source NoSQL systems and discuss some potential future research directions resulting from the survey.
-
Abstract
A major obstacle to developing artificial intelligence applications capable of true lifelong learning is that artificial neural networks quickly or catastrophically forget previously learned tasks when trained on a new one. Numerous methods for alleviating catastrophic forgetting are currently being proposed, but differences in evaluation protocols make it difficult to directly compare their performance. To enable more meaningful comparisons, here we identified three distinct scenarios for continual learning based on whether task identity is known and, if it is not, whether it needs to be inferred. Performing the split and permuted MNIST task protocols according to each of these scenarios, we found that regularization-based approaches (e.g., elastic weight consolidation) failed when task identity needed to be inferred. In contrast, generative replay combined with distillation (i.e., using class probabilities as "soft targets") achieved superior performance in all three scenarios. Addressing the issue of efficiency, we reduced the computational cost of generative replay by integrating the generative model into the main model by equipping it with generative feedback or backward connections. This Replay-through-Feedback approach substantially shortened training time with no or negligible loss in performance. We believe this to be an important first step towards making the powerful technique of generative replay scalable to real-world continual learning applications.
-
Abstract
Motivated by the widespread adoption of large-scale A/B testing in industry, we propose a new experimentation framework for the setting where potential experiments are abundant (i.e., many hypotheses are available to test), and observations are costly; we refer to this as the experiment-rich regime. Such scenarios require the experimenter to internalize the opportunity cost of assigning a sample to a particular experiment. We fully characterize the optimal policy and give an algorithm to compute it. Furthermore, we develop a simple heuristic that also provides intuition for the optimal policy. We use simulations based on real data to compare both the optimal algorithm and the heuristic to other natural alternative experimental design frameworks. In particular, we discuss the paradox of power: high-powered classical tests can lead to highly inefficient sampling in the experiment-rich regime.
-
Abstract
In many sequential decision making tasks, it is challenging to design reward functions that help an RL agent efficiently learn behavior that is considered good by the agent designer. A number of different formulations of the reward-design problem, or close variants thereof, have been proposed in the literature. In this paper we build on the Optimal Rewards Framework of Singh et.al. that defines the optimal intrinsic reward function as one that when used by an RL agent achieves behavior that optimizes the task-specifying or extrinsic reward function. Previous work in this framework has shown how good intrinsic reward functions can be learned for lookahead search based planning agents. Whether it is possible to learn intrinsic reward functions for learning agents remains an open problem. In this paper we derive a novel algorithm for learning intrinsic rewards for policy-gradient based learning agents. We compare the performance of an augmented agent that uses our algorithm to provide additive intrinsic rewards to an A2C-based policy learner (for Atari games) and a PPO-based policy learner (for Mujoco domains) with a baseline agent that uses the same policy learners but with only extrinsic rewards. Our results show improved performance on most but not all of the domains.
-
Abstract
Evolution strategies (ES) are a family of black-box optimization algorithms able to train deep neural networks roughly as well as Q-learning and policy gradient methods on challenging deep reinforcement learning (RL) problems, but are much faster (e.g. hours vs. days) because they parallelize better. However, many RL problems require directed exploration because they have reward functions that are sparse or deceptive (i.e. contain local optima), and it is unknown how to encourage such exploration with ES. Here we show that algorithms that have been invented to promote directed exploration in small-scale evolved neural networks via populations of exploring agents, specifically novelty search (NS) and quality diversity (QD) algorithms, can be hybridized with ES to improve its performance on sparse or deceptive deep RL tasks, while retaining scalability. Our experiments confirm that the resultant new algorithms, NS-ES and two QD algorithms, NSR-ES and NSRA-ES, avoid local optima encountered by ES to achieve higher performance on Atari and simulated robots learning to walk around a deceptive trap. This paper thus introduces a family of fast, scalable algorithms for reinforcement learning that are capable of directed exploration. It also adds this new family of exploration algorithms to the RL toolbox and raises the interesting possibility that analogous algorithms with multiple simultaneous paths of exploration might also combine well with existing RL algorithms outside ES.
-
Abstract
Direct representation knowledgebases can enhance and even provide an alternative to document-centered digital libraries. Here we consider realist semantic modeling of everyday activities and infrastructures in such knowledgebases. Because we want to integrate a wide variety of topics, a collection of ontologies (a foundry) and a range of other knowledge resources are needed. We first consider modeling the routine procedures that support human activities and technologies. Next, we examine the interactions of technologies with aspects of social organization. Then, we consider approaches and issues for developing and validating explanations of the relationships among various entities.
-
Abstract
We show how faceted search using a combination of traditional classification systems and mixed-membership topic models can go beyond keyword search to inform resource discovery, hypothesis formulation, and argument extraction for interdisciplinary research. Our test domain is the history and philosophy of scientific work on animal mind and cognition. The methods can be generalized to other research areas and ultimately support a system for semi-automatic identification of argument structures. We provide a case study for the application of the methods to the problem of identifying and extracting arguments about anthropomorphism during a critical period in the development of comparative psychology. We show how a combination of classification systems and mixed-membership models trained over large digital libraries can inform resource discovery in this domain. Through a novel approach of "drill-down" topic modeling---simultaneously reducing both the size of the corpus and the unit of analysis---we are able to reduce a large collection of fulltext volumes to a much smaller set of pages within six focal volumes containing arguments of interest to historians and philosophers of comparative psychology. The volumes identified in this way did not appear among the first ten results of the keyword search in the HathiTrust digital library and the pages bear the kind of "close reading" needed to generate original interpretations that is the heart of scholarly work in the humanities. Zooming back out, we provide a way to place the books onto a map of science originally constructed from very different data and for different purposes. The multilevel approach advances understanding of the intellectual and societal contexts in which writings are interpreted.
-
Abstract
In many real-world scenarios, rewards extrinsic to the agent are extremely sparse, or absent altogether. In such cases, curiosity can serve as an intrinsic reward signal to enable the agent to explore its environment and learn skills that might be useful later in its life. We formulate curiosity as the error in an agent's ability to predict the consequence of its own actions in a visual feature space learned by a self-supervised inverse dynamics model. Our formulation scales to high-dimensional continuous state spaces like images, bypasses the difficulties of directly predicting pixels, and, critically, ignores the aspects of the environment that cannot affect the agent. The proposed approach is evaluated in two environments: VizDoom and Super Mario Bros. Three broad settings are investigated: 1) sparse extrinsic reward, where curiosity allows for far fewer interactions with the environment to reach the goal; 2) exploration with no extrinsic reward, where curiosity pushes the agent to explore more efficiently; and 3) generalization to unseen scenarios (e.g. new levels of the same game) where the knowledge gained from earlier experience helps the agent explore new places much faster than starting from scratch. Demo video and code available at https://pathak22.github.io/noreward-rl/
-
Abstract
Attempts to train a comprehensive artificial intelligence capable of solving multiple tasks have been impeded by a chronic problem called catastrophic forgetting. Although simply replaying all previous data alleviates the problem, it requires large memory and even worse, often infeasible in real world applications where the access to past data is limited. Inspired by the generative nature of hippocampus as a short-term memory system in primate brain, we propose the Deep Generative Replay, a novel framework with a cooperative dual model architecture consisting of a deep generative model ("generator") and a task solving model ("solver"). With only these two models, training data for previous tasks can easily be sampled and interleaved with those for a new task. We test our methods in several sequential learning settings involving image classification tasks.
-
Abstract
Optimal stopping problems consider the question of deciding when to stop an observation-generating process in order to maximize a return. We examine the problem of simultaneously learning and planning in such domains, when data is collected directly from the environment. We propose GFSE, a simple and flexible model-free policy search method that reuses data for sample efficiency by leveraging problem structure. We bound the sample complexity of our approach to guarantee uniform convergence of policy value estimates, tightening existing PAC bounds to achieve logarithmic dependence on horizon length for our setting. We also examine the benefit of our method against prevalent model-based and model-free approaches on 3 domains taken from diverse fields.
-
Abstract
We start with an overview of a class of submodular functions called SCMMs (sums of concave composed with non-negative modular functions plus a final arbitrary modular). We then define a new class of submodular functions we call {\em deep submodular functions} or DSFs. We show that DSFs are a flexible parametric family of submodular functions that share many of the properties and advantages of deep neural networks (DNNs). DSFs can be motivated by considering a hierarchy of descriptive concepts over ground elements and where one wishes to allow submodular interaction throughout this hierarchy. Results in this paper show that DSFs constitute a strictly larger class of submodular functions than SCMMs. We show that, for any integer $k>0$, there are $k$-layer DSFs that cannot be represented by a $k'$-layer DSF for any $k'<k$. This implies that, like DNNs, there is a utility to depth, but unlike DNNs, the family of DSFs strictly increase with depth. Despite this, we show (using a "backpropagation" like method) that DSFs, even with arbitrarily large $k$, do not comprise all submodular functions. In offering the above results, we also define the notion of an antitone superdifferential of a concave function and show how this relates to submodular functions (in general), DSFs (in particular), negative second-order partial derivatives, continuous submodularity, and concave extensions. To further motivate our analysis, we provide various special case results from matroid theory, comparing DSFs with forms of matroid rank, in particular the laminar matroid. Lastly, we discuss strategies to learn DSFs, and define the classes of deep supermodular functions, deep difference of submodular functions, and deep multivariate submodular functions, and discuss where these can be useful in applications.
-
Abstract
CRDTs are distributed data types that make eventual consistency of a distributed object possible and non ad-hoc. Specifically, state-based CRDTs ensure convergence through disseminating the entire state, that may be large, and merging it to other replicas; whereas operation-based CRDTs disseminate operations (i.e., small states) assuming an exactly-once reliable dissemination layer. We introduce Delta State Conflict-Free Replicated Data Types ($\delta$-CRDTs) that can achieve the best of both worlds: small messages with an incremental nature, as in operation-based CRDTs, disseminated over unreliable communication channels, as in traditional state-based CRDTs. This is achieved by defining delta mutators to return a delta-state, typically with a much smaller size than the full state, that to be joined with both local and remote states. We introduce the $\delta$-CRDT framework, and we explain it through establishing a correspondence to current state-based CRDTs. In addition, we present an anti-entropy algorithm for eventual convergence, and another one that ensures causal consistency. Finally, we introduce several $\delta$-CRDT specifications of both well-known replicated datatypes and novel datatypes, including a generic map composition.
-
Abstract
We introduce Divnet, a flexible technique for learning networks with diverse neurons. Divnet models neuronal diversity by placing a Determinantal Point Process (DPP) over neurons in a given layer. It uses this DPP to select a subset of diverse neurons and subsequently fuses the redundant neurons into the selected ones. Compared with previous approaches, Divnet offers a more principled, flexible technique for capturing neuronal diversity and thus implicitly enforcing regularization. This enables effective auto-tuning of network architecture and leads to smaller network sizes without hurting performance. Moreover, through its focus on diversity and neuron fusing, Divnet remains compatible with other procedures that seek to reduce memory footprints of networks. We present experimental results to corroborate our claims: for pruning neural networks, Divnet is seen to be notably superior to competing approaches.
-
Abstract
Experience replay lets online reinforcement learning agents remember and reuse experiences from the past. In prior work, experience transitions were uniformly sampled from a replay memory. However, this approach simply replays transitions at the same frequency that they were originally experienced, regardless of their significance. In this paper we develop a framework for prioritizing experience, so as to replay important transitions more frequently, and therefore learn more efficiently. We use prioritized experience replay in Deep Q-Networks (DQN), a reinforcement learning algorithm that achieved human-level performance across many Atari games. DQN with prioritized experience replay achieves a new state-of-the-art, outperforming DQN with uniform replay on 41 out of 49 games.
-
Abstract
CRDTs are distributed data types that make eventual consistency of a distributed object possible and non ad-hoc. Specifically, state-based CRDTs ensure convergence through disseminating the en- tire state, that may be large, and merging it to other replicas; whereas operation-based CRDTs disseminate operations (i.e., small states) assuming an exactly-once reliable dissemination layer. We introduce Delta State Conflict-Free Replicated Datatypes ({\delta}-CRDT) that can achieve the best of both worlds: small messages with an incremental nature, as in operation-based CRDTs, disseminated over unreliable communication channels, as in traditional state-based CRDTs. This is achieved by defining {\delta}-mutators to return a delta-state, typically with a much smaller size than the full state, that is joined to both: local and remote states. We introduce the {\delta}-CRDT framework, and we explain it through establishing a correspondence to current state-based CRDTs. In addition, we present an anti-entropy algorithm that ensures causal consistency, and we introduce two {\delta}-CRDT specifications of well-known replicated datatypes.
-
Abstract
Determinantal point processes (DPPs), which arise in random matrix theory and quantum physics, are natural models for subset selection problems where diversity is preferred. Among many remarkable properties, DPPs offer tractable algorithms for exact inference, including computing marginal probabilities and sampling; however, an important open question has been how to learn a DPP from labeled training data. In this paper we propose a natural feature-based parameterization of conditional DPPs, and show how it leads to a convex and efficient learning formulation. We analyze the relationship between our model and binary Markov random fields with repulsive potentials, which are qualitatively similar but computationally intractable. Finally, we apply our approach to the task of extractive summarization, where the goal is to choose a small subset of sentences conveying the most important information from a set of documents. In this task there is a fundamental tradeoff between sentences that are highly relevant to the collection as a whole, and sentences that are diverse and not repetitive. Our parameterization allows us to naturally balance these two characteristics. We evaluate our system on data from the DUC 2003/04 multi-document summarization task, achieving state-of-the-art results.
-
Abstract
The Fedora architecture is an extensible framework for the storage, management, and dissemination of complex objects and the relationships among them. Fedora accommodates the aggregation of local and distributed content into digital objects and the association of services with objects. This al-lows an object to have several accessible representations, some of them dy-namically produced. The architecture includes a generic RDF-based relation-ship model that represents relationships among objects and their components. Queries against these relationships are supported by an RDF triple store. The architecture is implemented as a web service, with all aspects of the complex object architecture and related management functions exposed through REST and SOAP interfaces. The implementation is available as open-source soft-ware, providing the foundation for a variety of end-user applications for digital libraries, archives, institutional repositories, and learning object systems.