MARS: Memory-Enhanced Agents with Reflective Self-improvement

2025-03-25

AI Summary

To address the limitations of large language models (LLMs) in long-term memory retention, constrained context windows, and sustained decision-making under dynamic environments, this paper proposes a tri-agent collaborative framework—comprising User, Assistant, and Verifier agents—that integrates reflective reasoning, iterative feedback-driven policy optimization, and forgetting-aware memory management. Crucially, it introduces, for the first time, the Ebbinghaus forgetting curve into LLM memory retrieval and update mechanisms. We further design a reflective self-improvement mechanism that unifies multi-task coordination and long-horizon information preservation. Experiments on multi-turn dialogue and complex planning benchmarks demonstrate significant improvements: task completion rate and cross-turn consistency increase by 12.6% over strong baselines, validating the framework’s effectiveness in sustaining coherent, adaptive behavior over extended interactions.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Multiagent Systems: Multiagent LearningCognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactionsUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSearch and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Large language models (LLMs) have made significant advances in the field of natural language processing, but they still face challenges such as continuous decision-making, lack of long-term memory, and limited context windows in dynamic environments. To address these issues, this paper proposes an innovative framework Memory-Enhanced Agents with Reflective Self-improvement. The MARS framework comprises three agents: the User, the Assistant, and the Checker. By integrating iterative feedback, reflective mechanisms, and a memory optimization mechanism based on the Ebbinghaus forgetting curve, it significantly enhances the agents capabilities in handling multi-tasking and long-span information.

Problem

• Addresses continuous decision-making challenges in LLMs
• Solves lack of long-term memory in dynamic environments
• Overcomes limited context windows for multi-tasking

Innovation

• Integrates iterative feedback for multi-tasking
• Uses reflective mechanisms for self-improvement
• Optimizes memory via Ebbinghaus forgetting curve