SAGE: Self-evolving Agents with Reflective and Memory-augmented Abilities

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—Self-evolving Agents with Reflective and Memory-augmented Abilities (\textbf{SAGE}). The SAGE 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. The agents, through self-evolution, can adaptively adjust strategies, optimize information storage and transmission, and effectively reduce cognitive load. We evaluate the performance of the SAGE framework on multiple benchmarks and long text tasks. Experimental results show that SAGE significantly improves model performance, achieving a 2.26X improvement on closed-source models and an improvement ranging from 57.7% to 100% on open-source models, with particularly notable effects on smaller models.

Method

SCORE framework overview

An illustration of the SAGE: a user provides a description and instance to the assistant with short-term (STM) and long-term (LTM) memory. The assistant performs observation, action, reflection, and output, which the checker reviews. The retention rate curve on the right illustrates memory decay over time, with a self-evolving loop guiding continued updates.

BibTeX

@misc{liang2024selfevolvingagentsreflectivememoryaugmented,
      title={Self-evolving Agents with reflective and memory-augmented abilities}, 
      author={Xuechen Liang and Meiling Tao and Yinghui Xia and Tianyu Shi and Jun Wang and JingSong Yang},
      year={2024},
      eprint={2409.00872},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2409.00872}, 
}