Meta-Cognitive Memory Policy Optimization for Long-Horizon LLM Agents
For LLM agents tackling long-horizon tasks, MMPO provides a fine-grained memory optimization method that improves reliability at scale, though it is an incremental improvement over existing RL-based approaches.
MMPO introduces a self-supervised proxy, Belief Entropy, to localize and penalize poor intermediate memory summaries in LLM agents, achieving 97.1% performance on long-horizon tasks with up to 1.75M-token contexts, outperforming existing methods.
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.