Causal Memory Policy: Making Memory Utility Identifiable by Intervening on Retrieval

Arman Behnam, Binghui Wang|October 1, 2026arXiv

Key Takeaway

Current memory systems can't measure the value of memories that are never retrieved.

Summary

Memory-augmented language models struggle to identify which memories are actually useful because some memories are never retrieved, making their value impossible to measure.

reasoning evaluation training

Key Terms

retrieval-pipeline inverse-propensity-weighting causal-identification memory-augmented-language-model positivity-violation