Learning How to Remember: A Meta-Cognitive Management Method for Structured and Transferable Agent Memory

2026-01-12

AI Summary

This work addresses the limitations of existing large language model agents in long-horizon decision-making, which rely on fixed memory representations and consequently exhibit poor generalization under distributional shifts. To overcome this, the authors propose Meta-Cognitive Memory Abstraction (MCMA), a novel framework that formulates memory management as a learnable metacognitive capability. MCMA decouples task execution from memory by freezing the task model and introducing a trainable memory collaborator, enabling the construction of a multi-level abstract memory hierarchy. A task-similarity-driven mechanism selectively reuses stored memories, and the memory collaborator is trained via Direct Preference Optimization. Experiments demonstrate that MCMA significantly outperforms multiple baselines on ALFWorld, ScienceWorld, and BabyAI, achieving superior performance in task completion, out-of-distribution generalization, and cross-task transfer.

Technology Category

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

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language model (LLM) agents increasingly rely on accumulated memory to solve long-horizon decision-making tasks. However, most existing approaches store memory in fixed representations and reuse it at a single or implicit level of abstraction, which limits generalization and often leads to negative transfer when distribution shift. This paper proposes the Meta-Cognitive Memory Abstraction method (MCMA), which treats memory abstraction as a learnable cognitive skill rather than a fixed design choice. MCMA decouples task execution from memory management by combining a frozen task model with a learned memory copilot. The memory copilot is trained using direct preference optimization, it determines how memories should be structured, abstracted, and reused. Memories are further organized into a hierarchy of abstraction levels, enabling selective reuse based on task similarity. When no memory is transferable, MCMA transfers the ability to abstract and manage memory by transferring the memory copilot. Experiments on ALFWorld, ScienceWorld, and BabyAI demonstrate substantial improvements in performance, out-of-distribution generalization, and cross-task transfer over several baselines.

Problem

• agent memory
• memory abstraction
• transfer learning
• distribution shift
• generalization

Innovation

• Meta-Cognitive Memory Abstraction
• Memory Copilot
• Direct Preference Optimization
• Hierarchical Memory Abstraction
• Cross-Task Transfer