In the first part of the talk, we consider the problem of decomposing human motions to primitives. As an extension to sparse coding, we formulated the problem as a tensor factorization problem with tensor group norm regularization over the primitives as well as other relevant constraints on activations and primitives. We demonstrate the effectiveness of our approach to learn interpretable representations of human motion from motion capture data, and show that our approach outperforms recently developed matching pursuit and sparse coding algorithms.

In the second part, we investigate the problem of recognizing gesture sequence in sign language. We develop a semi-Markov conditional model approach, where feature functions are defined over segments of video and thus more flexible than linear-chain conditional model. We use neural network classifiers of letters and linguistic handshape features, along with expected motion profiles, to define segmental feature functions, and our model outperforms prior hidden Markov model baselines.

Taehwan Kim is a 5th year Ph.D candidate at Toyota Technological Institute at Chicago, advised by Karen Livescu. He received master in Computer Science at USC and bachelor in Computer Science & Engineering and Mathematics at POSTECH. His main research interests are machine learning and its application to time series data analysis such as human motion and gesture recognition.