Tag
This thread explores how the brain's dual memory systems (hippocampus and neocortex) offer lessons for building long-running AI agents, arguing that agents need a fast episodic capture and slow consolidation mechanism to avoid catastrophic interference, rather than relying solely on frozen models with temporary scaffolding.
HOLA (Hippocampal Linear Attention) augments linear attention with a bounded exact KV cache inspired by hippocampal memory, improving long-range recall and perplexity without sacrificing efficiency. At 340M parameters, it outperforms full-attention Transformers on Wikitext and achieves robust needle recall up to 32k tokens.
Auto-Dreamer introduces a learned offline memory consolidation method for language agents, decoupling fast memory acquisition from slow cross-session consolidation, and achieving higher performance with smaller memory banks, generalizing to unseen environments.
Introduces Cognifold, a brain-inspired always-on proactive memory for LLM agents that continuously organizes fragmented event streams into self-emerging cognitive structures via graph-topology self-organization, extending Complementary Learning Systems theory with a prefrontal intent layer.