Distributed Collaborative Activity Recognition

Yang’s group has been actively working in the field of activity recognition based on mobile devices. Their work includes transfer learning for activity recognition and activity recognition for indoor daily-living activity recognition via mobile devices (see http://www.cse.ust.hk/~qyang/byarea.htm#Activity_Recognition) at location, action and goal levels. Yang was an invited speaker at IJCAI Conference in 2009 to give a talk on activity recognition where he gave an overview of the dynamic research field (http://videolectures.net/ijcai09_yang_fllshli/). One of the systems developed by Yang’s group can transform the received sensor signals to predicted higher-level user actions and goals such as ‘taking a bus’, ‘turning on the stove’, etc. Then, once the action sequences are partially known, the system can also infer even higher-level user goals to be achieved, even when these goals are interleaving and concurrent. Using detected activities. Yang’s group gave an overview of their work in a video and via a software system known as VTrack. Recently, they have applied transfer learning to activity recognition in a system known as cross-domain activity recognition. Yang and his collaborators have designed a system that combines collaborative filtering in recommender systems, GPS and various other sources of data, into an integrated intelligent activity recognition system to recommend activities and venues for visitors in a large city. His work also covers mobile phone based abnormal activity recognition and activity profiling for the elderly, in order to provide e-health support. He has also been working with his students on the topic of collaborative filtering and large-scale cloud-based computation. In this area, Yang and his students have been active in research on recommendation systems and Web mining. They also developed a collaborative filtering approach that addresses the item-ranking problem directly by modeling user preferences derived from the ratings, and a novel transfer-learning methods for knowledge transfer between different collaborative filtering tasks (see http://www.cse.ust.hk/~qyang)