BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:Europe/Vienna
BEGIN:DAYLIGHT
DTSTART:20180325T030000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20181028T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260406T042328Z
UID:5b7695d7ad676976068014@ist.ac.at
DTSTART:20180905T100000
DTEND:20180905T110000
DESCRIPTION:Speaker: Michael Berry\nhosted by Gasper Tkacik\nAbstract: I wi
 ll present an overview of recent work from my lab suggesting that a common
  principle for population codes might be that neural activity is automatic
 ally organized into a discrete set of clusters. The argument is primarily 
 based on the analysis of populations of retinal ganglion cells using two d
 ifferent approaches. Using a hidden Markov model (HMM) that can accurately
  capture the statistics of population activity\, we find each latent state
  is a well-separated cluster. When we repeat the same visual stimulus\, we
  find that neural activity patterns are highly variable but that they map 
 onto just one or a few clusters. Thus\, clusters serve as population codew
 ords that exhibit error correction. In addition\, the receptive field of e
 ach cluster is qualitatively different than receptive fields of its consti
 tuent neurons. Thus\, clusters form a different basis set of visual featur
 es. Finally\, we show that there exist simple\, biologically plausible dec
 oding algorithms that can readout cluster identity. We will describe how t
 hese properties can be combined together to constitute a system of hierarc
 hical pattern recognition.Using a maximum entropy model (MaxEnt)\, we show
  that the retinal population is in a low temperature state\, analogous to 
 a spin glass in statistical physics\, where each basin in the energy lands
 cape is a cluster of neural activity. Because the properties of any MaxEnt
  model depend entirely on the constraining statistics (such as firing rate
 s and pairwise correlations)\, the above results may also apply to neural 
 populations elsewhere in the brain that have the same low-order statistics
 . To this end\, we show that if we scale down all of observed pairwise cor
 relations by a factor >2\, the retina remains in a low temperature state. 
 This suggests that other population codes with pairwise correlations as st
 rong as the retina may also exist in a low temperature limit\, where neura
 l activity is organized into clusters.  We present preliminary evidence th
 at neural populations in primary visual cortex are also organized into clu
 sters.
LOCATION:Mondi Seminar Room 3\, Central Building\, ISTA
ORGANIZER:abonvent@ist.ac.at
SUMMARY:Michael Berry: A Design Principle for Population Neural Codes
URL:https://talks-calendar.ista.ac.at/events/1352
END:VEVENT
END:VCALENDAR