HeLa-Mem: Hebbian Learning and Associative Memory for LLM Agents

arXiv cs.CL / 4/21/2026

📰 NewsDeveloper Stack & InfrastructureModels & Research

共有:

Key Points

The paper highlights that LLM agents struggle with long-term coherence because fixed context windows and embedding-based retrieval often miss the associative structure of human memory.
HeLa-Mem introduces a bio-inspired memory architecture using Hebbian learning, representing memories as a dynamic graph governed by mechanisms akin to association, consolidation, and spreading activation.
It uses a dual-level design: an episodic memory graph that updates from co-activation patterns, and a semantic store built via “Hebbian Distillation” by a Reflective Agent that extracts structured knowledge from dense memory hubs.
Experiments on LoCoMo show HeLa-Mem outperforms baselines across four question categories while using significantly fewer context tokens, and the authors provide an open-source GitHub repository.

Abstract

Long-term memory is a critical challenge for Large Language Model agents, as fixed context windows cannot preserve coherence across extended interactions. Existing memory systems represent conversation history as unstructured embedding vectors, retrieving information through semantic similarity. This paradigm fails to capture the associative structure of human memory, wherein related experiences progressively strengthen interconnections through repeated co-activation. Inspired by cognitive neuroscience, we identify three mechanisms central to biological memory: association, consolidation, and spreading activation, which remain largely absent in current research. To bridge this gap, we propose HeLa-Mem, a bio-inspired memory architecture that models memory as a dynamic graph with Hebbian learning dynamics. HeLa-Mem employs a dual-level organization: (1) an episodic memory graph that evolves through co-activation patterns, and (2) a semantic memory store populated via Hebbian Distillation, wherein a Reflective Agent identifies densely connected memory hubs and distills them into structured, reusable semantic knowledge. This dual-path design leverages both semantic similarity and learned associations, mirroring the episodic-semantic distinction in human cognition. Experiments on LoCoMo demonstrate superior performance across four question categories while using significantly fewer context tokens. Code is available on GitHub: https://github.com/ReinerBRO/HeLa-Mem