发表机构
Shanghai Academy of AI for Science; Fudan University; Shanghai Jiao Tong University; University of Michigan; The Chinese University of Hong Kong; Alibaba Group; Nanjing University; Shanghai Innovation Institute（上海人工智能科学研究院; 复旦大学; 上海交通大学; 密歇根大学; 香港中文大学; 阿里巴巴集团; 南京大学; 上海创新研究院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究提出REMORY神经记忆网络，通过补充软记忆令牌辅助冻结LLM处理长视野任务，在SummHay等基准测试中提升源归因、减少工具错误，仅用5.2%输入位置接近全上下文性能。
AI 中文摘要
长视野智能体需压缩历史信息以在有限上下文窗口内运行，但仅靠文本摘要可能无法支持后续所有决策。本文提出REMORY，一种神经记忆网络，它用有界序列的软记忆令牌补充摘要。给定历史信息和摘要，该网络学习生成令牌，帮助冻结的大语言模型（LLM）近似其使用完整历史信息会生成的后续内容。这些令牌以摘要为条件，附加在摘要之后，形成序列维度上残差连接的类似物。在SummHay数据集上，REMORY在洞察覆盖率几乎不变的情况下提升了源归因，且仅使用5.2%的输入位置就接近了全上下文联合得分。在长视野智能体基准测试中，Qwen3.8-27B和GLM-5.3-Flash在使用残差记忆后均取得一致提升，且在BrowseComp和Terminal-Bench 2.1上的工具输出重复次数和工具错误也显著减少。
英文摘要
Long-horizon agents compact their history to continue within a finite context window, but a textual summary alone may not support every subsequent decision. We introduce REMORY, a neural memory network that supplements the summary with a bounded sequence of soft memory tokens. Given the history and summary, the network learns to generate tokens that help a frozen LLM approximate the continuation it would produce with the full history. The tokens are conditioned on the summary and appended after it, forming an analogue of a residual connection along the sequence dimension. On SummHay, REMORY improves source attribution at nearly unchanged insight coverage and approaches the full-context joint score using only 5.2% of the input positions. Across long-horizon agent benchmarks, Qwen3.8-27B and GLM-5.3-Flash show consistent gains with residual memory. Both models also exhibit substantially fewer repeated tool outputs and tool errors on BrowseComp and Terminal-Bench 2.1.
