Biologically-Inspired Computational Models of Vision

The theme for our May event is Biologically-Inspired Computational Models of Vision. Food will be provided by Affectiva (http://www.affectiva.com/)! Here's an outline of the agenda:

7:00-7:30 pm: Networking session. 7:30-8:05 pm: "Visual recognition and visual search in the human brain: bottom-up models and beyond", Dr. Gabriel Kreiman (http://klab.tch.harvard.edu/index.html)
8:05-8:40 pm: "The computational magic of pattern recognition in cortex: A theory of selectivity and invariance", Dr. Tomaso Poggio (http://cbcl.mit.edu/cbcl/)

_____________________________________________

Visual recognition and visual search in the human brain: bottom-up models and beyond

Gabriel Kreiman
Children's Hospital, Harvard Medical School

Abstract: The rapid cascade of processes leading up to visual recognition has been successfully described by purely bottom-up architectures (e.g. Tomaso Poggio’s work). These biologically-inspired, hierarchical and bottom-up computational models also display a rather remarkable performance in computer vision tasks, particularly in cases that involve identifying or categorizing isolated objects or objects embedded in images with small amounts of clutter. To begin to investigate the limits of feed-forward processing and possible additional advantages conferred by the incorporation of feed-back signals to such models, we considered two everyday visual tasks: searching for objects in cluttered images and recognizing partially occluded objects. We characterized human performance in these tasks at the behavioral level and we capitalized on a rare opportunity to peek inside the human brain to examine the brain areas and dynamics in visual cortex during those tasks. The behavioral and physiological observations suggest significant processing delays when locating a target object or identifying an object from a fraction of the pixels. These delays may represent the need for additional computational processing including feedback and recurrent loops. I will discuss demonstrations of computational architectures that extend the bottom-up models by incorporating feedback and which can improve performance in visual search and recognition of occluded objects.

Biography: Gabriel Kreiman is Assistant Professor in the Department of Ophthalmology at Children’s Hospital Boston, Harvard Medical School and Principal Investigator at the Kreiman Lab. He received his MSc and PhD degrees from the California Institute of Technology (Caltech) and subsequently worked as a research fellow at the Massachusetts Institute of Technology. He received the 2009 National Institutes of Health New Innovator Award and the 2010 National Science Foundation Career Award.

_____________________________________________

The computational magic of pattern recognition in cortex: A theory of selectivity and invariance

Tomaso Poggio
CBCL, McGovern Institute, Massachusetts Institute of Technology

Abstract: I conjecture that the sample complexity of object recognition is mostly due to geometric image transformations and that a main goal of the ventral stream is to learn-and-discount image transformations while preserving sufficient selectivity. The theory predicts that the size of the receptive fields determines which transformations are learned during development; that the transformation represented in each area determines the tuning of the neurons in the area; and that class-specific transformations are learned and represented at the top of the ventral stream hierarchy. In problems of pattern recognition, hierarchical, layered architectures -- similar to cortex -- may exploit in an optimal way unsupervised learning of transformations to provide invariant and discriminative signatures to a supervised classifier.

Biography: Tomaso A. Poggio, is the Eugene McDermott Professor in the Dept. of Brain & Cognitive Sciences at MIT and a member of both the Computer Science and Artificial Intelligence Laboratory and of the McGovern Institute. He is
an honorary member of the Neuroscience Research Program, a member of the American Academy of Arts and Sciences, a Founding Fellow of AAAI, a founding member of the McGovern Institute for Brain Research. Among other honors, he received the Laurea Honoris Causa from the University of Pavia for the Volta Bicentennial, the 2003 Gabor Award, the Okawa Prize 2009, and the AAAS Fellowship. He is one of the most cited computational Scientists (h-index=116, according to GoogleScholar) with contributions ranging from the biophysical and behavioral studies of the visual system to the computational analyses of vision and learning in humans and machines. With W. Reichardt he characterized quantitatively the visuomotor control system in the y. With D. Marr, he introduced the seminal idea of levels of analysis in computational neuroscience. He introduced regularization as a mathematical framework to approach the ill-posed problems of vision and the key problem of learning from data. He has contributed to the early development of the theory of learning -- in particular introducing concepts such as RBFs, supervised learning in RKHSs and stability as a necessary and sufficient condition for generalization. In the last decade he has developed an influential quantitative model of visual recognition in the visual cortex. The citation for the recent 2009 Okawa prize mentions his "... outstanding contributions to the establishment of computational neuroscience, and pioneering researches ranging from the biophysical and behavioral studies of the visual system to the computational analysis of vision and learning in humans and machines."