Search papers, labs, and topics across Lattice.
This paper introduces Metacognitive Memory Policy Optimization (MMPO), a novel approach to training memory-augmented LLM agents that focuses on optimizing the clarity of the agent's belief state based on intermediate memory summaries. MMPO uses a self-supervised proxy called Belief Entropy to penalize summaries that induce high epistemic uncertainty about the latent task state. Experiments demonstrate that MMPO significantly outperforms existing methods on long-horizon tasks, maintaining high performance even with very long contexts (1.75M tokens).
Don't just reward success; penalize memory summaries that make your LLM agent uncertain about the task at hand.
Memory-augmented LLM agents tackle complex long-horizon tasks by recursively summarizing interaction trajectories into compact memory. However, existing approaches typically train these memory policies using outcome-based reinforcement learning, failing to localize where intermediate memory quality degrades. As interactions unfold, ambiguous recursive summaries progressively discard task-relevant information and introduce semantic noise. This exacerbates belief deviation, obscuring the agent's estimate of the latent task state and ultimately derailing long-horizon reasoning. We therefore argue that memory optimization should focus not merely on trajectory-level success, but on the clarity of the belief induced by intermediate summaries. To this end, we introduce Belief Entropy, a self-supervised proxy that probes how uncertain the model remains about the latent task state given its current memory. Based on this proxy, we propose Metacognitive Memory Policy Optimization (MMPO). Instead of relying only on sparse outcome-based signals, MMPO provides fine-grained, memory-specific supervision via explicitly penalizing summaries that induce high epistemic uncertainty. Experiments show that MMPO consistently outperforms existing methods on diverse long-horizon tasks, maintaining 97.1% performance even when scaled to 1.75M-token contexts.
