Events & News
Colloquium
| Category | Lecture |
| Date | Wednesday, April 8, 2009 |
| Time | 11:00 am |
| Concludes | 12:00 pm |
| Location | SU 411-A |
| Speaker | Rich Zemel, U. Toronto |
Learning to label complex images
Image parsing entails the interpretation of visual images, forming descriptions of the scene depicted in the image. One step towards this goal involves assigning to each pixel or region of an image one of a predefined set of labels (e.g., road, tree, sky). This image labeling task is an increasingly popular problem in the machine learning and machine vision communities. In this talk I will describe the most successful approaches to this goal, which involve including contextual information in the process. For example, a local image patch may be ambiguous as to whether it contains a hippopotamus or a polar bear;
however, a nearby patch containing snow and ice can help resolve this ambiguity. We have developed methods for representing and learning such contextual information, and demonstrated their utility in labeling complex real-world images. I will also describe recent approaches that can benefit from training images with partial or noisy labels.
Biography
Richard Zemel was born in Montreal and grew up in Pittsburgh. He attended
Harvard University, where he received a B.A. degree in History and Science,
focusing on Applied Mathematics and American History. He worked at Carnegie
Group for a few years. He then attended the Department of Computer Science at the University of Toronto, where he obtained his Ph.D. in 1994, with U.S. Alumni and NSERC Fellowships. His dissertation is titled "A Minimum Description Length Framework for Unsupervised Learning", and his supervisor was Dr. Geoff Hinton.
Rich then moved to San Diego for a couple years, where he was a postdoc in the Computational Neurobiology Laboratory at the Salk Institute. He then became a postdoc in the Department of Psychology at Carnegie Mellon University, supported by a McDonnell-Pew Cognitive Neuroscience Postdoctoral Fellowship. He then moved to Tucson, where he held faculty appointments in the Department of Psychology and the Department of Computer Science at the University of Arizona. Then in the fall of 2000, he took a leave of absence from Arizona and moved back to Toronto, becoming a faculty member in the Department of Computer Science at the
University of Toronto.
His research interests cover a range of topics in machine learning, visual
perception, and neural coding. Specific interests include unsupervised learning,boosting, perceptual learning, representations of visual motion, multisensory integration, and probabilistic models of neural representations.