跨模态可解释性个体行为分析与健康管理

对个体行为和健康状态进行实时、连续的分析和监测，有助于实现在线健康管理、预警和干预，提高个人健康管理能力,形成健康生活方式，预防和减少疾病发生。本项目面向基于智能手机、穿戴式设备等采集的实时数据流研究可解释性的个体行为分析和健康管理，探索解决多源异构数据流的统一表征、高冗余数据个体差异敏感的行为分析、动态健康因素关联挖掘与可靠性决策三个关键科学问题，重点研究跨模态时序数据关联与融合、缺失特征补全与全息表示、少样本行为识别、个体自适应行为序列检测、动静态结合的健康状态预测和隐式健康因素分析与行为推荐。最终建立多源异构个体行为识别与推荐方面的基础理论方法，完成安全、可信、有因可循的一体化技术解决方案。通过开发健康行为管理助手APP，将研究成果集成到智能手机在实际应用环境下形成示范应用，有助于加快人工智能的创新研究与应用，推动健康管理实现从点状监测向连续监测、从短流程管理向长流程管理转变。

The real-time and continuous analysis and monitoring of individual behaviors and health statuses are conducive to realize the online health management, early warning, and intervention, to improve the personal health management ability, to form a healthy lifestyle, to prevent and reduce the occurrence of diseases. Based on the real-time data stream collected by smartphones and wearable devices, this project focuses on the interpretable individual behavior analysis and health management. The three principle scientific problems are the unified representation of the multi-source heterogeneous data stream, the individual-sensitive behavior analysis on the highly redundant data stream, the dynamic health factor mining and reliable decision making. The research program includes the correlation and fusion of the cross-modal sequential data, the missing feature completion and holographic feature learning, few-shot behavior recognition, individual-adaptive behavior sequence detection, dynamic-static health status prediction, implicit health factor analysis and behavior recommendation. Finally, the theoretical basics of individual behavior recognition and recommendation on the multi-source heterogeneous data stream will be established. A secure, credible and causal integrated technical solution will be achieved. Through the development of health behavior management assistant APP, the research achievement will be integrated into smartphones to form a pilot application in the real environment. This project will help to accelerate the innovative research and application of artificial intelligence. It will also promote the transformation of health management from spot monitoring to continuous monitoring and from short process management to long process management.