← 首页|学术|GraphMemix: Query-Aware Evidence Forests for Long-Term Multimodal Agent Memory
cs.AI · 2608.26983 · 2026-08-27

GraphMemix: Query-Aware Evidence Forests for Long-Term Multimodal Agent Memory

Geng Li, Yuhao Wang, Dong Li, Jianye Hao, Yuxin Peng
多模态记忆Agent Memory长期记忆
💬 多模态 agent 的长期记忆不该是「离线摘要」或「粗暴相似度检索」,而应该按查询构建一片「证据森林」。

🎯 背景

现有多模态 agent 长期记忆组织方式要么是昂贵的、与问题无关的离线摘要,要么是朴素的 embedding 相似度匹配,前者浪费计算,后者容易引入不完整或冗余的上下文。

🔬 方法

GraphMemix 把记忆组织建模为「查询感知的证据森林构建」组合优化问题,包含三个组件:(1) 候选图构建——通过 schema 和语义关系扩展多视角种子记忆,获取查询相关的原始上下文;(2) 证据效用与激活成本——把直接记忆支持和锚点条件的关系验证解耦,抑制冗余或冲突信息;(3) 森林优化——在最大证据预算下联合选择森林格式的记忆上下文及其可靠的关系结构。

📊 结果与意义

在四个长期多模态记忆基准上相较不同基础模型都有显著提升,并在准确率与生命周期成本之间建立了新的帕累托前沿——这正是用户关注的「多模态记忆架构」核心方向的新进展。
▶ 原文摘要 Abstract
Organizing long-term memory for multimodal agents remains challenging because existing methods either suffer from expensive question-agnostic offline summaries or naive embedding similarity matching that introduces incomplete and redundant context. To address these issues, we propose GraphMemix, a combinatorial-optimization graph memory framework that models memory organization as query-aware evidence-forest construction. Specifically, our method consists of three key components:(1) candidate graph construction, which expands multi-view seed memories through schema and semantic relations to acquire query-aware original context; (2) evidence utility and activation costs, which decouples direct memory support from anchor-conditioned relation verification to suppress redundant or conflicting information; and (3) forest optimization, which jointly selects a forest-format memory context under a maximum evidence budget and its reliable relational structure. By organizing memory into a query-relevant subgraph, the method avoids substantial lifecycle cost and recovers low-similarity complementary evidence. Experimental results across four long-term multimodal memory benchmarks demonstrate significant improvements with different foundation models and establish a new Pareto frontier between accuracy and lifecycle cost.
来源:arXiv:2608.26983 · 精读基于摘要与 arXiv HTML/abs 页信息生成,未解析 PDF 全文