Abstract:

Recently, the computer vision community has made impressive progress on object recognition with deep learning approaches. However, for any visual system to interact with objects, it needs to understand much more than simply recognizing where the objects are. In fact, this would require studying the representations and algorithms necessary for a system to physically interact in everyday scenes. In this talk, I will describe how we can develop deep learning models with inspiration from the way humans learn. More specifically, my research explores and solves how to obtain interaction skills through observation and interaction over time. It will cover several recent deep learning techniques such as generative adversarial network, reinforcement learning, and attention models.

Biography:

Joseph Lim is an assistant professor of computer science at the University of Southern California (USC). Previously, he was a postdoctoral scholar with Fei-Fei Li at Stanford University. Before that, he completed his PhD at Massachusetts Institute of Technology under the guidance of Antonio Torralba, and also had a half-year long postdoc with William Freeman. He received his undergraduate degree at the University of California - Berkeley, where he worked with Jitendra Malik. His research interests are in computer vision and machine learning. He is particularly interested in deep learning, reinforcement learning, structure learning, and multi-modal data, especially for teaching machines how to interact with their physical surroundings.

All faculties and students are welcome to join.