CraniMem (arXiv:2603.15642): Cranial-Inspired Gated Memory Beats Baselines on Long-Horizon Agent Recall

arXiv·medium signal

CraniMem proposes a neurocognitively motivated multi-stage memory architecture for agents: goal-conditioned attentional gating suppresses low-salience inputs before encoding, a bounded episodic buffer handles near-term continuity, and a structured long-term knowledge graph stores durable semantic recall. A scheduled consolidation loop replays high-utility traces into the graph while pruning low-utility items, keeping memory growth bounded. CraniMem consistently outperforms baselines on robustness to distractors and long-horizon stability — the primary failure mode for production agent deployments.