BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Swiss Computational Neuroscience Seminar: Haim Sompolinsky
DTSTART:20181122T160500
DTEND:20181122T165500
DTSTAMP:20260512T015615Z
UID:e1e58d5044bfa323f9f3c0a7b7f4b355a88caaa4549ff4d50a18dec3
CATEGORIES:Conferences - Seminars
DESCRIPTION:Haim Sompolinsky\nTitle: Untangling of Perceptual Manifolds i
 n Deep Networks\nAbstract: An object perceived under different physical c
 onditions (i.e. location\, pose\, size\, orientation\,\nbackground) create
 s different responses in sensory neurons\, resulting in a “perceptual\nm
 anifold” in the response-space of a neuronal population. A prominent the
 ory asserts that\nhierarchical sensory systems untangle these manifolds\, 
 allowing downstream systems to\nperform perceptually invariant tasks such 
 as object recognition and classification. However\,\nit is unclear who to 
 quantify the process of untangling and measure it in biological and\nartif
 icial deep networks. In my talk\, I will describe recent theoretical advan
 ces that relate the\nability to perform invariant object classification to
  the geometric properties of the perceptual\nmanifolds. I will show numeri
 cal results that use the theory to evaluate the population based\nchanges 
 of object representations in the successive layers of deep convolutional n
 euronal\nnetworks (DCNNs) as well as in neural data in visual cortex. This
  work contributes to the\nconstruction of a systematic theory of sensory p
 rocessing in deep networks in both AI and\nthe brain.
LOCATION:SV 1717 https://plan.epfl.ch/?room==SV%201717
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR