Memp: Exploring Agent Procedural Memory
This work addresses the issue of brittle procedural memory for LLM-based agents, offering a method to improve task performance and efficiency, though it is incremental in nature.
The paper tackles the problem of brittle procedural memory in LLM-based agents by proposing Memp, a learnable and updatable memory system that distills past trajectories into fine-grained instructions and script-like abstractions. Empirical results show steadily higher success rates and greater efficiency on tasks like TravelPlanner and ALFWorld, with memory migration to weaker models yielding substantial performance gains.
Large Language Models (LLMs) based agents excel at diverse tasks, yet they suffer from brittle procedural memory that is manually engineered or entangled in static parameters. In this work, we investigate strategies to endow agents with a learnable, updatable, and lifelong procedural memory. We propose Memp that distills past agent trajectories into both fine-grained, step-by-step instructions and higher-level, script-like abstractions, and explore the impact of different strategies for Build, Retrieval, and Update of procedural memory. Coupled with a dynamic regimen that continuously updates, corrects, and deprecates its contents, this repository evolves in lockstep with new experience. Empirical evaluation on TravelPlanner and ALFWorld shows that as the memory repository is refined, agents achieve steadily higher success rates and greater efficiency on analogous tasks. Moreover, procedural memory built from a stronger model retains its value: migrating the procedural memory to a weaker model yields substantial performance gains.