Home  ›  Resource  ›  Lectures  ›  Content

Pattern-match decision making categorization: what can Machine Vision learn from Neuroscience

Title: Seminar: Pattern-match decision making categorization: what can Machine Vision learn from Neuroscience

Speaker: Xiao-Jing Wang

         Department of Neurobiology Kavli Institute for Neuroscience,

         Director of Swartz Program in Theoretical Neurobiology,

         Yale University School of Medicine,

         Chair Professor of Neural Cognitive Computation,

         TNList, Tsinghua University

Venue: FIT 小报告厅(一楼大厅后侧)

Time: 16:00-17:00, May 26 (Wednesday), 2010

Host: Prof. Bo Zhang, Department of Computer Science & Technology


The interaction between Machine Vision Neuroscience is poised to be fruitful, but progress depends on fresh ideas new approaches. I will argue that it is time to go beyond early visual processing, tackle visual cognition such as decision-making categorization. In this general informal talk, I will illustrate how we do this type of research using neural circuit modeling. I will first discuss “pattern match”. The ability to judge whether sensory stimuli match an internally represented pattern is central to many brain functions. To elucidate the underlying mechanism, we developed a neural circuit model for match/nonmatch decision-making. At the core of this model are two neural populations that show match enhancement suppression as a result of inhibition-dominated recurrent dynamics heterogeneous top-down excitation from a working memory circuit. A downstream system extracts the necessary information to make match/nonmatch decisions. The model acs for key observations in several delayed match-to-sample experiments, including a task where irrelevant stimulus repetitions (e.g. BB in ABBA) must be ignored. A testable prediction is that the magnitudes of match enhancement suppression are parametrically tuned to the similarity between two compared patterns. Furthermore, reward-dependent synaptic plasticity enables decision neurons to flexibly adjust the readout scheme according to task dems, whereby the most informative neural signals have the highest impact.



Xiao-Jing Wang is Professor of Neurobiology, adjunct professor of Physics Psychology, director of the Swartz Program in Theoretical Neurobiology at Yale University. He obtained his Ph. D. degree in Theoretical Physics at the University of Brussels in 1987, when he changed his field of research to Computational Neuroscience. He uses theory biophysically realistic neural circuit modeling to study the brain mechanisms of cognitive functions, especially decision-making. He has published over 90 papers in Nature Neuroscience, Proceedings of the National Academy of Sciences, Neuron other esteemed journals.