Event Location: NSH 1507
Bio: Jia Xu is a PhD candidate of the Computer Sciences Department at the University of Wisconsin-Madison, working with Prof. Vikas Singh, Prof. Jerry Zhu, and Prof. Chuck Dyer. He was a visiting student at University of Toronto in Summer 2014 and at TTI-Chicago during Summer 2013, both working with Prof. Raquel Urtasun. Jia obtained a B.S in Computer Science and Technology from Nanjing University in 2010. His major research interests are in computer vision, machine learning, and optimization, with a particular focus on visual parsing.

Abstract: Despite promising performance from conventional fully supervised algorithms, visual parsing has remained an important yet challenging problem, due to the limited size of fully annotated data available, and/or high cost to collect such annotations. Therefore, it is of great interest to come up with solutions that can learn to segment from weak human supervision or weakly labeled data, as it is much cheaper to collect or readily available at much larger scale. In this talk, I will present our current efforts towards visual parsing under weak supervision, with a particular focus on addressing the following scientific questions:

When human comes into the loop, how can we minimize user efforts while still achieving satisfactory parsing results?

How can we utilize weakly labeled data effectively into the visual parsing task?

Once we have such a reliable visual parsing machinery in hand, how can we interpret/organize massive visual data more efficiently, as well as helping the aging and/or disabled population (with Alzheimer Disease, Autism, Blindness etc.)?