Contextual Experience Replay for Self-Improvement of Language Agents

• URL: http://arxiv.org/abs/2506.06698v1
• Date: Sat, 07 Jun 2025 07:47:35 GMT
• Title: Contextual Experience Replay for Self-Improvement of Language Agents
• Authors: Yitao Liu, Chenglei Si, Karthik Narasimhan, Shunyu Yao,
• Abstract summary: We propose Contextual Experience Replay (CER) to enable efficient self-improvement for language agents.<n>CER accumulates and synthesizes past experiences into a dynamic memory buffer.<n>We evaluate CER on the challenging WebArena and VisualWebArena benchmarks.
• Score: 47.51006612841945
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language model (LLM) agents have been applied to sequential decision-making tasks such as web navigation, but without any environment-specific experiences, they often fail in these complex tasks. Moreover, current LLM agents are not designed to continually learn from past experiences during inference time, which could be crucial for them to gain these environment-specific experiences. To address this, we propose Contextual Experience Replay (CER), a training-free framework to enable efficient self-improvement for language agents in their context window. Specifically, CER accumulates and synthesizes past experiences into a dynamic memory buffer. These experiences encompass environment dynamics and common decision-making patterns, allowing the agents to retrieve and augment themselves with relevant knowledge in new tasks, enhancing their adaptability in complex environments. We evaluate CER on the challenging WebArena and VisualWebArena benchmarks. On VisualWebArena, CER achieves a competitive performance of 31.9%. On WebArena, CER also gets a competitive average success rate of 36.7%, relatively improving the success rate of the GPT-4o agent baseline by 51.0%. We also conduct a comprehensive analysis on it to prove its efficiency, validity and understand it better.

Related papers

• SWE Context Bench: A Benchmark for Context Learning in Coding [6.093520696434546]
  We introduce SWE-ContextBench, a benchmark designed to explicitly evaluate experience reuse in programming agents.<n>Built on SWE-Bench Lite, SWE-ContextBench augments 300 base tasks with 99 related tasks derived from real dependency and reference relationships among GitHub issues and pull requests.<n>We show that correctly selected summarized experience improves resolution accuracy and substantially reduces runtime and token cost.
  arXiv  Detail & Related papers  (2026-02-09T06:44:45Z)
• LANPO: Bootstrapping Language and Numerical Feedback for Reinforcement Learning in LLMs [73.27182315028021]
  LANPO is a framework that cleanly separates the roles of feedback: language guides exploration, while numerical rewards drive optimization.<n>Our work provides a robust method for integrating historical experiences into the LLM RL loop, creating more effective and data-efficient learning agents.
  arXiv  Detail & Related papers  (2025-10-18T15:51:19Z)
• ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization [86.57743637722596]
  Large Language Model (LLM)-based web agents demonstrate strong performance on knowledge-intensive tasks.<n>Complex queries involving multiple entities, intertwined relationships, and high uncertainty demand extensive search cycles.<n>We introduce ReSum, a novel paradigm that enables indefinite exploration through periodic context summarization.
  arXiv  Detail & Related papers  (2025-09-16T17:57:22Z)
• MCP-Universe: Benchmarking Large Language Models with Real-World Model Context Protocol Servers [86.00932417210477]
  We introduce MCP-Universe, the first comprehensive benchmark specifically designed to evaluate LLMs in realistic and hard tasks through interaction with real-world MCP servers.<n>Our benchmark encompasses 6 core domains spanning 11 different MCP servers: Location Navigation, Repository Management, Financial Analysis, 3D Design, Browser Automation, and Web Searching.<n>We find that even SOTA models such as GPT-5 (43.72%), Grok-4 (33.33%) and Claude-4.0-Sonnet (29.44%) exhibit significant performance limitations.
  arXiv  Detail & Related papers  (2025-08-20T13:28:58Z)
• Teaching Language Models To Gather Information Proactively [53.85419549904644]
  Large language models (LLMs) are increasingly expected to function as collaborative partners.<n>In this work, we introduce a new task paradigm: proactive information gathering.<n>We design a scalable framework that generates partially specified, real-world tasks, masking key information.<n>Within this setup, our core innovation is a reinforcement finetuning strategy that rewards questions that elicit genuinely new, implicit user information.
  arXiv  Detail & Related papers  (2025-07-28T23:50:09Z)
• REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites [9.58858258192147]
  We introduce REAL, a benchmark and framework for multi-turn agent evaluations on deterministic simulations of real-world websites.<n>We also release a benchmark consisting of 112 practical tasks that mirror everyday complex user interactions.<n>Our framework supports easy integration of new tasks, reproducible evaluation, and scalable post-training data generation.
  arXiv  Detail & Related papers  (2025-04-15T18:22:55Z)
• Learn-by-interact: A Data-Centric Framework for Self-Adaptive Agents in Realistic Environments [33.83610929282721]
  Learn-by-interact is a data-centric framework to adapt large language models (LLMs) to any given environments without human annotations.<n>We assess the quality of our synthetic data by using them in both training-based scenarios and training-free in-context learning (ICL)<n>Experiments on SWE-bench, WebArena, OSWorld and Spider2-V spanning across realistic coding, web, and desktop environments show the effectiveness of Learn-by-interact.
  arXiv  Detail & Related papers  (2025-01-18T22:34:41Z)
• The BrowserGym Ecosystem for Web Agent Research [151.90034093362343]
  BrowserGym ecosystem addresses the growing need for efficient evaluation and benchmarking of web agents.<n>We propose an extended BrowserGym-based ecosystem for web agent research, which unifies existing benchmarks from the literature.<n>We conduct the first large-scale, multi-benchmark web agent experiment and compare the performance of 6 state-of-the-art LLMs across 6 popular web agent benchmarks.
  arXiv  Detail & Related papers  (2024-12-06T23:43:59Z)
• Evaluating Cultural and Social Awareness of LLM Web Agents [113.49968423990616]
  We introduce CASA, a benchmark designed to assess large language models' sensitivity to cultural and social norms.<n>Our approach evaluates LLM agents' ability to detect and appropriately respond to norm-violating user queries and observations.<n>Experiments show that current LLMs perform significantly better in non-agent environments.
  arXiv  Detail & Related papers  (2024-10-30T17:35:44Z)
• ExACT: Teaching AI Agents to Explore with Reflective-MCTS and Exploratory Learning [78.42927884000673]
  ExACT is an approach to combine test-time search and self-learning to build o1-like models for agentic applications.<n>We first introduce Reflective Monte Carlo Tree Search (R-MCTS), a novel test time algorithm designed to enhance AI agents' ability to explore decision space on the fly.<n>Next, we introduce Exploratory Learning, a novel learning strategy to teach agents to search at inference time without relying on any external search algorithms.
  arXiv  Detail & Related papers  (2024-10-02T21:42:35Z)
• Large Language Models Can Self-Improve At Web Agent Tasks [37.17001438055515]
  Large language models (LLMs) have recently demonstrated some capability to navigate novel environments as agents in a zero-shot or few-shot fashion. We explore the extent to which LLMs can self-improve their performance as agents in long-horizon tasks in a complex environment using the WebArena benchmark. We achieve a 31% improvement in task completion rate over the base model on the WebArena benchmark through a self-improvement procedure.
  arXiv  Detail & Related papers  (2024-05-30T17:52:36Z)
• ERAGent: Enhancing Retrieval-Augmented Language Models with Improved Accuracy, Efficiency, and Personalization [14.62114319247837]
  We introduce ERAGent, a cutting-edge framework that embodies an advancement in the RAG area. Retrieval Trigger is incorporated to curtail extraneous external knowledge retrieval without sacrificing response quality. ERAGent personalizes responses by incorporating a learned user profile.
  arXiv  Detail & Related papers  (2024-05-06T04:42:18Z)
• ExpeL: LLM Agents Are Experiential Learners [57.13685954854463]
  We introduce the Experiential Learning (ExpeL) agent to allow learning from agent experiences without requiring parametric updates.<n>Our agent autonomously gathers experiences and extracts knowledge using natural language from a collection of training tasks.<n>At inference, the agent recalls its extracted insights and past experiences to make informed decisions.
  arXiv  Detail & Related papers  (2023-08-20T03:03:34Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.