基于细粒度迁移的跨场景认知能力评估方法  

A fine-grained transfer learning method for cross-scenario cognitive-ability assessment

作　　者：张迎伟[1,2] 陈益强 于汉超[4] 杨晓东 谷洋[1,2,3] Yingwei ZHANG;Yiqiang CHEN;Hanchao YU;Xiaodong YANG;Yang GU(Beijing Key Laboratory of Mobile Computing and Pervasive Devices,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China;Peng Cheng Laboratory,Shenzhen 518066,China;Bureau of Frontier Sciences and Education,Chinese Academy of Sciences,Beijing 100864,China)

机构地区：[1]中国科学院计算技术研究所移动终端与新型计算北京市重点实验室,北京100190 [2]中国科学院大学,北京100049 [3]鹏城实验室,深圳518066 [4]中国科学院前沿科学与教育局,北京100864

出　　处：《中国科学：信息科学》2023年第4期647-662,共16页Scientia Sinica(Informationis)

基　　金：广东省重点领域研发计划(批准号:2019B010109001);国家自然科学基金(批准号:61972383,61902377,61902379);山东省智研院创新课题(批准号:SDAICT2191010);中国科学院青年创新促进会资助。

摘　　要：认知健康是大脑健康的重要组成部分,与个体全生命周期的正常发展密切相关.目前,非受限普适场景下的认知能力评估已成为实现认知障碍相关疾病早期预警的重要途径.然而,在医疗、家庭等不同计算场景下,因感知设备、认知评估流程等的不同,往往存在不同场景间评估模型相互适用难的问题.针对以上跨场景认知能力评估挑战,本文提出了一种参数自适应的细粒度迁移学习方法 PAFG-TL.PAFG-TL基于随机森林模型实现,通过参数自适应的个体分类器评估策略和领域自适应的决策树生长机制实现参数无关的个体分类器聚类评估和决策树细粒度进化生长.通过在临床认知评估和公开基准数据集上进行实验验证,证明了PAFG-TL方法在跨场景认知能力评估中的有效性.Cognitive health is an important aspect of brain health and is closely related to an individual’s development throughout their life.Currently,cognitive-ability assessment in unrestricted ubiquitous scenarios has gained attention for realizing early warnings of cognitive impairment-related diseases.However,the perception devices and differences in the cognitive assessment processes in different application scenarios(e.g.,hospital and home)influence cross-scenario usage.To address these challenges,we propose a parameter-adaptive,fine-grained transfer-learning method(PAFG-TL),which is designed based on the random forest algorithm.It is composed of a parameter adaptation-based model-evaluation strategy and a domain-adaptive tree-growing mechanism to realize a parameter-independent clustering evaluation of individual classifiers and the fine-grained evolutionary growth of decision trees.In this study,the experimental verification of clinical cognitive assessment and public benchmark datasets demonstrated the effectiveness of PAFG-TL in cross-scenario cognitive-ability assessment.

关 键 词：参数自适应 细粒度 迁移学习 跨场景 认知能力评估 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] R749.1[自动化与计算机技术—控制科学与工程]