Validation-Gated Hebbian Learning for Adaptive Agent Memory

Pragya Singh, Stanley Yu

Keywords: LLM Agents, Hebbian Learning, Knowledge Graphs, Multi-Agent Systems, Adaptive Memory, Neuroplasticity-Inspired AI, Neural-Symbolic AI, Hallucination Mitigation

Abstract: LLM-based agents struggle with catastrophic forgetting, context limitations, and reasoning drift. While knowledge graphs (KGs) offer structured memory, current implementations remain static and do not adapt based on reasoning effectiveness. We introduce Kairos, a multi-agent reasoning system implementing Hebbian plasticity mechanisms for adaptive knowledge graphs. Kairos formalizes three neuroplasticity-inspired operations: edge strengthening (LTP analog), temporal decay (LTD analog), and emergent connection formation. A key innovation is validation-gated learning, where graph consolidation only occurs when reasoning passes multi-dimensional quality assessment (logical, grounding, novelty, alignment), preventing hallucination reinforcement. Our controlled proof-of-concept demonstrates that validation-gated Hebbian learning is both mechanically sound and practically beneficial, with adaptive graphs outperforming static baselines. We additionally identify a design principle showing that novelty and correctness are orthogonal dimensions that degrade when averaged in validation systems. These results establish the feasibility of adaptive agent memory where knowledge structures improve through validated iteration.

Submission Number: 14