AdMem:基于预期落地的程序化记忆框架,让长期任务智能体越用越强
AdMem: Expectation-Grounded Procedural Memory for Long-Horizon Task-Solving Agents
这篇论文把"智能体记忆为什么会帮倒忙"讲得很透:朴素存经验反而掉分,AdMem 用预期-评分配对加上效用值驱动的遗忘机制解决。做 agent 记忆系统的开发者直接能借鉴它的信用分配设计,全文记忆还是可读文本,方便调试审计。
arXiv 论文提出 AdMem,一个面向长期任务智能体的程序化记忆框架。论文首先给出可复现的负面观察:在 AgentBoard 基准上,朴素的工作流归纳记忆在 8 个领域中的 7 个里,解题数反而少于无记忆的 ReAct 智能体;在 Jericho 环境中,无结构的逐步程序化记忆库会损失 ReAct 一半的完成量。
AdMem 补上两个关键机制:一是按动作分配信用——执行前 actor 先承诺预期结果,并行的 critic 将实际结果与预期对比,写出带二元奖励和反思的程序化条目,无需环境奖励信号;二是给记忆条目明确的生命周期,每条记忆携带一个标量效用值,在噪声-OR 奖励模型下在线更新,与上下文相似度共同决定检索排序,并驱动遗忘与合并,避免记忆库无限膨胀。
在 AgentBoard 八个领域共 707 个任务上,AdMem 解出 441 个,ReAct 为 388 个,工作流归纳基线仅 312 个,合并收益通过双侧 Fisher 精确检验显著,且增益集中在具备跨任务可迁移结构的领域。所有记忆保持人类可读文本,使失败可诊断,但论文也指出这种可读存储同时构成可审计却可被攻击的攻击面。
AdMem: Expectation-Grounded Procedural Memory for Long-Horizon Task-Solving Agents
An agent that carries memory across tasks is only as good as the entries it keeps. We begin from an observation that is easy to reproduce and uncomfortable for the field: naively accumulating cross-task experience can make an agent \emph{worse} than one with no memory at all. On AgentBoard, a workflow-induction memory solves fewer tasks than a memoryless ReAct agent in seven of eight domains, and a per-step procedural store with no surrounding structure loses half of ReAct's completions on Jericho. We argue that the missing ingredients are per-action credit assignment and an explicit life cycle for stored entries, and present AdMem, which supplies both. Before each action the actor commits to an expected outcome; a concurrently running critic scores the realized outcome against that expectation and writes a procedural entry carrying a binary reward and a reflection, so credit lands on the action that earned it and no environment reward is required. Each entry then carries a scalar utility, updated online under a noisy-OR reward model, that ranks retrieval jointly with context similarity and drives eviction and merging, giving the store a forgetting and consolidation policy instead of unbounded growth. Across 707 tasks in eight AgentBoard domains AdMem solves 441 against 388 for ReAct and 312 for the workflow-induction baseline --- a pooled gain significant under a two-sided Fisher exact test, concentrated in the domains that supply transferable cross-task structure. All memory stays human-readable text, which is what makes the failures above diagnosable --- and, as we discuss, also makes the store an auditable but attackable surface.
本条情报归属 AITOP「论文」核心栏目,重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:AgentBoard、AdMem、ReAct、智能体记忆,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明:本文原始线索采编自「arXiv cs.AI」(@Runzhe Wang, Huilin Lu, Shengjie Liu, Li Dong, Zhihuai Zhu),经 AITOP 多维评分管线结构化提炼。