Thesis Defense: Minjoon Kouh - "Toward a More Biologically Plausible Model of Object Recognition"
Speaker: Minjoon Kouh , Physics Department , MIT
Date: May 10 2007
Time: 1:00PM to 3:00PM
Location: Rm. 46-3310
Host: Prof. Tomas Poggio, BCS Dept., McGovern Institute and CSAIL
Contact: Mary Pat Fitzgerald, 617-253-0551, marypat@csail.mit.edu
Relevant URL: Rapidly and reliably recognizing an object (is that a cat or a tiger?) is obviously an important skill for survival. However, it is a difficult computational problem, because the same object can appear differently under various conditions, while different objects can share similar features. An emerging picture from decades of neuroscience research is that the cortex overcomes this challenge by gradually building up specificity and invariance with a hierarchical architecture.
I will present an extended version of a computational model of object recognition, originally proposed by Riesenhuber and Poggio (1999). The model quantitatively describes the anatomy, physiology, and the first few hundred milliseconds of visual information processing in the ventral pathway of the primate visual cortex. I will show that the two main operations in the model (Gaussian and maximum) can be cast into a more biologically plausible form, using monotonic nonlinearities and divisive normalization. I will propose a possible canonical neural circuitry that can compute these operations and is closely related to other models of cortical nonlinearities (e.g., energy model and other divisive normalization models). Second, I will show that the shape tuning properties of the model units are in good agreements with several experimental data measured from the corresponding visual areas, in particular V4 and the inferotemporal cortex.
See other events that are part of Brains & Machines Seminar Series 2007
See other events happening in May 2007
