{"id":"arxiv:2601.03192","title":"MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory","date":"2026-01-06","section":"Agents, open-endedness, AGI","quality":0.457268,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/","source_url":"https://arxiv.org/abs/2601.03192","text_available":true,"authors":["Shengtao Zhang","Jiaqian Wang","Ruiwen Zhou","Junwei Liao"],"outgoing":[{"source":"arxiv:2601.03192","target":"arxiv:2503.14476","evidence":"id","id":"arxiv:2503.14476","title":"DAPO: An Open-Source LLM Reinforcement Learning System at Scale","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/548e934b912ef84691ace63bf1cfd670e7b0257d9d6986dde926f569e7b2f8b7/"},{"source":"arxiv:2601.03192","target":"doi:10.1073/pnas.1611835114","evidence":"title","id":"doi:10.1073/pnas.1611835114","title":"Overcoming catastrophic forgetting in neural networks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/849082368a44d06b93ef3ba95bcfddc6672b7a5e8064ad2b2d19601639266e8b/"},{"source":"arxiv:2601.03192","target":"url:selfrag.github.io","evidence":"id","id":"url:selfrag.github.io","title":"Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/f2a6b92a158b280a6699d4fda0ed474777be788abf9ed561ab7d61da1da57ad1/"}],"incoming":[{"source":"arxiv:2604.08224","target":"arxiv:2601.03192","evidence":"id","id":"arxiv:2604.08224","title":"Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/392317a8b3bedec8258ff8288d69ac72bc9e60299c5f7bec8962838bb448c2ce/"},{"source":"arxiv:2607.13104","target":"arxiv:2601.03192","evidence":"id","id":"arxiv:2607.13104","title":"Self-Improvements in Modern Agentic Systems: A Survey","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/"},{"source":"arxiv:2608.19197","target":"arxiv:2601.03192","evidence":"id","id":"arxiv:2608.19197","title":"SPADE: Self-Play in Adaptive Synthetic Executable Environments","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c524376d6034b5d79661d6f1a444591404f4a250bcb0790d802c05e02f8249df/"},{"source":"arxiv:2609.37633","target":"arxiv:2601.03192","evidence":"id","id":"arxiv:2609.37633","title":"RLTL;DR: Self-improvement by Internalizing Self-generated Feedback","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c7a4c1dbb0493059883d9a01c1f89d4db3644209b909e1195c5afa4d3a320999/"}],"text_parts":[{"start":0,"end":15690,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html"},{"start":15690,"end":31467,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html"},{"start":31467,"end":46815,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html"},{"start":46815,"end":62688,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html"},{"start":62688,"end":78041,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html"},{"start":78041,"end":81243,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html"}],"contents":[{"title":"MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"Abstract","offset":85,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"Keywords:","offset":1142,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"1 Introduction","offset":1160,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"2 Related Works","offset":5344,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"Runtime Learning","offset":5364,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"Reinforcement Learning","offset":6494,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"Agentic Memory","offset":7568,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"3 Problem Formulation","offset":8926,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"3.1 Memory-Augmented Agent Policy","offset":9446,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"3.2 Non-Parametric Reinforcement Learning","offset":11391,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"4 MemRL","offset":14377,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"4.1 The Intent-Experience-Utility Triplet","offset":15028,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"4.2 From Semantic Recall to Value-Aware Selection","offset":15635,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-1.html#text"},{"title":"4.3 Non-Parametric RL on Memory","offset":17435,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"4.4 Theoretical Stability Analysis","offset":18571,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5 Experiments","offset":20488,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.1 Experimental Setup","offset":20506,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.2 Main Results","offset":21720,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"Runtime Learning Results.","offset":21742,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"Transferring Results.","offset":22322,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.3 Ablations","offset":22925,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.3.1 Effectiveness of Runtime RL","offset":22944,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.3.2 Impact of Q-Value Weighting","offset":23585,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.3.3 Ablation Analysis: Cross-Task vs. Single-Task Optimization","offset":24329,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.3.4 Sensitivity to Retrieval Size ( k 1 k_{1} and k 2 k_{2} ).","offset":25411,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.4 Discussion","offset":26261,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.4.1 Predictive Power of the Q Critic","offset":26460,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.4.2 Stability of MemRL","offset":27473,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"5.4.3 Extended Analysis.","offset":28858,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"6 Limitations and Conclusion","offset":29454,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"Limitations and Future Work.","offset":29487,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"Conclusion.","offset":30639,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"Impact Statement","offset":31446,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-2.html#text"},{"title":"Appendix A Theoretical Analysis and Proofs","offset":32374,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.1 Stability Analysis","offset":32421,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Setup.","offset":32859,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Stationary Reward Assumption.","offset":33612,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Expected Convergence of Utility Estimates.","offset":34299,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Theorem A.1 .","offset":34390,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Proof.","offset":34989,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Bounded Variance and Stability.","offset":35835,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Global Stability via EM Convergence.","offset":36496,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.2 Proof of Theorem A.1 : Convergence of EMA Estimation","offset":38737,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Assumptions.","offset":39029,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Derivation of Error Dynamics.","offset":39558,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Convergence Analysis.","offset":40458,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.3 Bounded Variance and Global Stability","offset":41803,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Derivation of the Variance Bound.","offset":42139,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Recursive Unrolling.","offset":43123,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Asymptotic Convergence.","offset":44050,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"Connection to Phase-A Clustering.","offset":44925,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.4 Convergence via Variational Inference","offset":45721,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.4.1 The Convergence Objective","offset":45940,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.4.2 Variational Objective with Trust Region","offset":46763,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-3.html#text"},{"title":"A.4.3 Optimization via Generalized Expectation-Maximization (GEM)","offset":48219,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"E-Step (Policy Optimization).","offset":48414,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"M-Step (Policy Evaluation via Error Minimization).","offset":52417,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"A.4.4 Proof of Convergence","offset":53873,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"Appendix B Extended Analysis and Insights","offset":54864,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"B.1 Detailed Analysis of Forgetting Dynamics","offset":54910,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"B.2 MemRL as a Trajectory Verifier.","offset":56017,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"B.3 Impact of Task Similarity on Memory Efficacy","offset":57008,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"Appendix C Advanced Capabilities: Transfer and Merging","offset":59350,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"C.1 Cross-Model Memory Transferability","offset":59409,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"C.2 Multi-Task Memory Merging and Interference Analysis","offset":60747,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"Appendix D Baseline and Benchmark Details","offset":62185,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"D.1 Baselines","offset":62529,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"I. Test-Time Scaling Strategy","offset":62651,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-4.html#text"},{"title":"II. Retrieval-Augmented Generation (RAG) Approaches","offset":63057,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"III. Agentic Memory Systems","offset":63794,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"D.2 Benchmark Datasets","offset":64418,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"Appendix E Implementation and Reproducibility Details","offset":66054,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"E.1 Model Specifications","offset":66264,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"E.2 Hyperparameter Settings","offset":66488,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"E.3 Data Partitioning","offset":66956,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"E.4 Model Selection and Performance Analysis","offset":67670,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"Appendix F Cost and Efficiency Analysis","offset":70932,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"F.1 Token Consumption","offset":71262,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"F.2 Runtime Latency and Stability","offset":72073,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"Negligible Algorithmic Overhead.","offset":72342,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"Appendix G Extended Limitations and Future Work","offset":73313,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"G.1 Update Protocols and Memory Consolidation","offset":73557,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"G.2 Precise Credit Assignment in Multi-Memory Updates","offset":74216,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"G.3 Task Similarity and Generalization","offset":74992,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"G.4 Memory Security and Robustness to Attack","offset":76298,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"G.5 Multi-Agent Collaboration and Shared Memory","offset":77159,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-5.html#text"},{"title":"G.6 Dedicated Domains and Hybrid Architectures","offset":78792,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html#text"},{"title":"Appendix H Case Study: High-Utility Failure Analysis (Near-Misses)","offset":79552,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html#text"},{"title":"Appendix I Prompts","offset":79913,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html#text"},{"title":"I.1 Experience Summarization Prompts","offset":80195,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html#text"},{"title":"I.2 Generation and Inference Prompts","offset":80237,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html#text"},{"title":"Instructions for reporting errors","offset":80279,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/part-6.html#text"}],"text_code_points":81243,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/source.md"}