JD - arXivDaily 大厂专区

2607.13940 2026-07-16 cs.AI 新提交

A Self-Evolving Agent for Longitudinal Personal Health Management

一种用于纵向个人健康管理的自我进化智能体

Haoran Li, Jiebi Deng, Tong Jin, Jinghong Han, Yuxin Wang, Zexin Wang, Qingyi Si, Weikang Gong, Xiahai Zhuang, Jia You, Wei Cheng, Jianfeng Feng, Hongcheng Guo

发表机构 * School of Data Science, Fudan University（复旦大学数据科学学院）； School of Life Sciences, Beijing University of Chinese Medicine（北京中医药大学生命科学学院）； Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University（复旦大学脑科学与智能技术研究院）； School of Computer Science and Technology, Huazhong University of Science and Technology（华中科技大学计算机科学与技术学院）； JD.com, Inc.（京东公司）； Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence, Fudan University, Ministry of Education（复旦大学计算神经科学与类脑智能教育部重点实验室）； Department of Neurology, Huashan Hospital, Fudan University（复旦大学附属华山医院神经内科）

AI总结研究针对多数健康AI系统孤立处理请求的问题，开发开源智能体架构HealthClaw，通过自我进化更新支持，经合成基准和生物医学任务评估，在准确率、隐私性及任务指标增益上表现出色，支持纵向个人健康智能体的自我进化记忆。

Comments 20 pages, 4 figures, 6 supplementary tables. Code: https://github.com/HC-Guo/HealthClaw

详情

AI中文摘要

个人健康管理是在反复接触中展开的，但大多数健康人工智能系统孤立地处理每个请求。我们开发了HealthClaw，这是一种开源智能体架构，它会随着个人日常、偏好、测量和风险的变化更新支持。它将共享安全规则和医学知识与包含个人资料事实、可重复使用程序和情景痕迹的私人纵向记忆分开。每次事件后，归纳法决定应更新个人资料、修订程序、保留情景还是排除。我们用合成的一年期基准和九个200例生物医学任务评估了HealthClaw。在900次纵向支持探测中，答案准确率从当前查询提示的0.2%提高到HealthClaw的45.7%，同时提示侧上下文暴露比全历史提示低71.7%。在100次隐私探测中，HealthClaw产生了更高的隐私感知答案质量和更少的不安全披露。在生物医学任务中，特定任务主要指标的平均绝对增益为27.0个百分点，经过错误发现率校正后，七个增益仍然显著。这些离线基准支持纵向个人健康智能体的受治理、自我进化记忆，尽管临床有效性需要前瞻性评估。HealthClaw可在这个https网址公开获取。

英文摘要

Personal health management unfolds over repeated encounters, yet most health AI systems treat each request in isolation. We developed HealthClaw, an open-source agent architecture that updates support as a person's routines, preferences, measurements and risks change. It separates shared safety rules and medical knowledge from private longitudinal memory containing profile facts, reusable procedures and episodic traces. After each episode, induction determines what should update the profile, revise a procedure, remain episodic or be excluded. We evaluated HealthClaw with a synthetic year-long benchmark and nine 200-case biomedical tasks. Across 900 longitudinal support probes, answer accuracy increased from 0.2% with current-query prompting to 45.7% with HealthClaw, while prompt-side context exposure was 71.7% lower than with full-history prompting. In 100 privacy probes, HealthClaw produced higher privacy-aware answer quality and fewer unsafe disclosures than both baselines. Across the biomedical tasks, the mean absolute gain in the task-specific primary metric was 27.0 percentage points, and seven gains remained significant after false-discovery-rate correction. These offline benchmarks support governed, self-evolving memory for longitudinal personal health agents, although clinical effectiveness requires prospective evaluation. HealthClaw is publicly available at https://github.com/HC-Guo/HealthClaw.

URL PDF HTML ☆