MemPrism：面向长程智能体的任务条件关系记忆视图
MemPrism: Task-Conditioned Relational Memory Views for Long-Horizon Agents
- Nanyang Technological University（南洋理工大学）
- South China University of Technology（华南理工大学）
- Beijing University of Posts and Telecommunications（北京邮电大学）
- University of Science and Technology of China（中国科学技术大学）
- Peking University（北京大学）
- Zhejiang University（浙江大学）
- Fudan University（复旦大学）
- Shanghai Jiao Tong University（上海交通大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
该研究针对长程智能体记忆系统的表示不匹配问题，提出MemPrism任务条件关系记忆框架，经实验验证可提升长程任务性能、减少记忆消耗且视图策略可跨VLM迁移。
AI中文摘要：
长程智能体依赖记忆来复用经验，但现有记忆系统通常假设证据可通过固定表示直接使用，这会导致表示不匹配——相关信息虽存在却未针对当前决策进行组织。为此，我们提出MemPrism，这是一种任务条件关系记忆框架，它将持久经验存储与决策时的工作记忆分离开。MemPrism将交互记录为事件流，并根据当前任务上下文动态构建关系视图；轻量视图策略会选择关系结构、证据范围、结果条件和粒度，而确定性组合器与渲染器则将历史事实转换为临时的光学工作记忆视图，供冻结的任务策略使用。在长程具身智能体和网络智能体基准上的实验表明，MemPrism在轨迹变长时始终提升任务性能，同时减少记忆令牌消耗；此外，学习到的视图策略无需额外适配即可跨不同视觉语言模型（VLM）迁移，证明了任务条件关系视图作为智能体通用记忆接口的有效性。
英文摘要：
Long-horizon agents rely on memory to reuse experiences, yet existing memory systems often assume that evidence can be directly consumed through a fixed representation. This leads to representation mismatch, where relevant information is available but not organized for the current decision. To this end, we propose MemPrism, a task-conditioned relational memory framework that separates persistent experience storage from decision-time working memory. MemPrism records interactions as the event stream and dynamically constructs relational views according to the current task context. A lightweight view policy selects the relation structure, evidence range, outcome condition, and granularity, while a deterministic composer and render transform historical facts into a temporary optical working-memory view for a frozen task policy. Experiments on long-horizon embodied and web-agent benchmarks show that MemPrism consistently improves the task performance, especially as trajectories become longer, while reducing memory token consumption. Furthermore, the learned view policy transfers across different VLMs without additional adaptation, demonstrating the effectiveness of task-conditioned relational views as a general memory interface for agents.
