October 12, 2021

12:00 pm / 1:00 pm

Venue

https://wse.zoom.us/j/99567504456?pwd=WkI2UlpGT3p6MldLS05VNkdmcGxiZz09

recorded seminar:
https://wse.zoom.us/rec/share/6heecAZNjeVmUukaGL_Z9dJfRLOZ429sFkh-3IXnjUBPpygy-PLTsZszOhlOD7E1.YQA-UrGjYLuY9MkX?startTime=1634054551000

BrunoOIshausen, PhD

Professor

Neuroscienceand Optometry

Universityof California, Berkeley

?Perception as Inference? 

Abstract:  Our subjective experience of thevisual worldis nothing like the 2D images sensed by our retinae.  Weexperience astable world full of well-defined geometric shapes and objectboundaries, 3D surfaces with continuous shading and reflectance along with theirmaterial properties, yet the retinal image is highly unstable, discretelysampledby a highly non-uniform lattice of retinal ganglion cells, and relayedto the cortex via punctate spike trains.  How do we explain this? Ishall present the point of view that our subjective experience is a mostlycorrect hallucination about the external world which the brain works very hardto create.  It may be understood in mechanistic terms by expanding uponHelmholtz’s notion of ?perception as inference’ using the mathematicalframework of Bayesian inference. I will describe three example of thisapproach:  a computational model for how the cortex achieves high-acuityvisual representations from fixational drift motion, the robustness of sparsecoding to adversarial perturbations, and a hierarchical Bayesian inferencemodel of visual cortex.

Biography: Bruno OIshausen is Professor of Neuroscienceand Optometry at the University of California, Berkeley.  He also servesas Director of the Redwood Center for Theoretical Neuroscience, aninterdisciplinary research group focusing on mathematical and computationalmodels of brain function.  He received B.S. and M.S. degrees in ElectricalEngineering from Stanford University, and a Ph.D. in Computation and NeuralSystems from the California Institute of Technology.  During postdoctoralwork with David Field at Cornell he developed the sparse coding model of visualcortex which provides a linking principle between natural scene statistics andthe response properties of visual neurons.  Olshausen’s current researchaims to understand the information processing strategies employed by the brainfor doing tasks such as object recognition and scene analysis.  This workseeks not only to advance our understanding of the brain, but also to discovernew algorithms for scene analysis based on how brains work.