{"ID":5438620,"CreatedAt":"2026-07-01T01:17:58.482524686Z","UpdatedAt":"2026-07-03T04:20:05.427450767Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2606.31121","arxiv_id":"2606.31121","title":"The Past Is Prologue: A Plug-in Controller for Selective Updates in Sequentially Evolving LLM Memory","abstract":"Sequentially evolving LLM memory enables agents to reuse past experience, but existing systems usually deploy each locally generated memory update without checking whether it improves future behavior. As a result, updates that help the current task may overwrite useful knowledge, introduce over-specific rules, or bias the final memory toward recent examples. We propose Janus, a plug-in memory controller that decides whether to accept a candidate memory update or retain the previous memory. To make this decision efficient, Janus uses a Memory Momentum Trigger to identify suspicious deviations in the memory-update trajectory, and compares old and new memories on a compact hybrid evaluation set of coverage, boundary, and fresh tasks instead of replaying the full history. Janus is method-agnostic and wraps existing updaters without changing their update rules. Across six datasets, two backbone LLMs, and two memory updaters, Janus improves average accuracy by +2.7 to +4.6 points over the corresponding base updaters.","short_abstract":"Sequentially evolving LLM memory enables agents to reuse past experience, but existing systems usually deploy each locally generated memory update without checking whether it improves future behavior. As a result, updates that help the current task may overwrite useful knowledge, introduce over-specific rules, or bias...","url_abs":"https://arxiv.org/abs/2606.31121","url_pdf":"https://arxiv.org/pdf/2606.31121v1","authors":"[\"Zihan Chen\",\"Songwei Dong\",\"Chengshuai Shi\",\"Peng Wang\",\"Song Wang\",\"Cong Shen\",\"Jundong Li\"]","published":"2026-06-30T04:33:18Z","proceeding":"cs.AI","tasks":"[\"cs.AI\"]","methods":"[\"Large Language Model\"]","has_code":false}