CSE Faculty Seminar: Honglak Lee, PhD

Title: New Directions in Deep Representation Learning Abstract: Deep learning methods have recently emerged as successful techniques to learn feature hierarchies from unlabeled and labeled data. In this talk, I will present my perspectives on the progress, challenges, and some new directions. Specifically, I will talk about my recent work to address the following interrelated challenges: (1) How can we learn invariant yet discriminative features, and furthermore disentangle underlying factors of variation to model high-order interactions between the factors and perform analogical reasoning? (2) How can we learn representations of the output data when the output variables have complex dependencies (such as in image segmentation/labeling and object detection)? (3) How can we learn shared representations from heterogeneous input data modalities (such as video and audio, image and text) with a theoretical guarantee? Bio: Honglak Lee is an Assistant Professor of Computer Science and Engineering at the University of Michigan, Ann Arbor. He received his PhD from the Computer Science Department at Stanford University in 2010, advised by Prof. Andrew Ng. His research focuses on deep learning and representation learning, which spans over unsupervised and semi-supervised learning, supervised learning, transfer learning, structured prediction, graphical models, and optimization. His methods have been successfully applied to computer vision and other perception problems. He received best paper awards at ICML and CEAS. He has served as a guest editor of IEEE TPAMI Special Issue on Learning Deep Architectures, as well as area chairs of ICML, NIPS, ICCV, AAAI, IJCAI, and ICLR. He received the Google Faculty Research Award (2011), NSF CAREER Award (2015), and was selected by IEEE Intelligent Systems as one of AI’s 10 to Watch (2013).