Adaptive Machine Learning of Interaction Sequences

Multivariate sequences are ubiquitious in interaction scenarios, e.g.   body gestures, speech, data streams from smart rooms or interface operation. This project develops methods to learn such data in highly dynamic environments,  such as companion-based interaction with smart rooms, in which unpredictable changes may occur and the systems have to re-adjust at many events.  We develop models for unsupervised and reinforcement learning of hierarchical structure from sequential data both for the analysis and generation of multivariate sequence patterns.