HeLa-Mem: Bio-Inspired Hebbian Memory Graph for LLM Agents
arXiv·medium signal
Models agent memory as a dynamic graph with Hebbian learning dynamics—memories strengthen through co-activation patterns. Dual-level organization: episodic memory graph evolves through use, while semantic memory store maintains structured knowledge. Biologically-inspired approach to the persistent memory problem that avoids the brittleness of fixed retrieval pipelines.