@inproceedings{hou-etal-2026-flashmem,
    title = "{F}lash{M}em: Distilling Intrinsic Latent Memory via Computation Reuse",
    author = "Hou, Yubo  and
      Chen, Zhisheng  and
      Wan, Tao  and
      Qin, Zengchang",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.230/",
    pages = "4687--4705",
    ISBN = "979-8-89176-395-1",
    abstract = "The stateless architecture of Large Language Models inherently lacks the mechanism to preserve dynamic context, compelling agents to redundantly reprocess history to maintain long-horizon autonomy. While latent memory offers a solution, current approaches are hindered by architectural segregation, relying on auxiliary encoders that decouple memory from the reasoning backbone. We propose \textbf{FlashMem}, a framework that distills intrinsic memory directly from transient reasoning states via computation reuse. Leveraging the property that internal representations uniquely encode input trajectories, FlashMem identifies the last hidden state as a sufficient statistic for the interaction history. This enables a \textbf{Shared-KV Consolidator} to synthesize memory by attending directly to the backbone{'}s frozen cache, eliminating redundant re-parameterization. Furthermore, a parameter-free \textbf{Cognitive Monitor} leverages attention entropy to adaptively trigger consolidation only when high epistemic uncertainty is detected. Experiments demonstrate that FlashMem matches the performance of heavy baselines while reducing inference latency by \textbf{5 times}, effectively bridging the gap between efficiency and persistent cognition."
}