BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Computational Neuroscience Journal Club - Adriana Perez-Rotondo (C
 ontrol group)
DTSTART:20181030T140000Z
DTEND:20181030T150000Z
UID:TALK114151@talks.cam.ac.uk
CONTACT:Rodrigo Echeveste
DESCRIPTION:This time exceptionally at *2pm*\n\nAdriana Perez-Rotondo will
  be presenting:\n\n• Sparseness and Expansion in Sensory Representations
 \n\n• Baktash Babadi and Haim Sompolinsky\n\n• Neuron (2014)\n\n• ht
 tps://www.cell.com/neuron/fulltext/S0896-6273(14)00646-1\n\n\nAbstract: In
  several sensory pathways\, input stimuli project to sparsely active downs
 tream populations that have more neurons than incoming axons. Here\, we ad
 dress the computational benefits of expansion and sparseness for clustered
  inputs\, where different clusters represent behaviorally distinct stimuli
  and intracluster variability represents sensory or neuronal noise. Throug
 h analytical calculations and numerical simulations\, we show that expansi
 on implemented by feed-forward random synaptic weights amplifies variabili
 ty in the incoming stimuli\, and this noise enhancement increases with spa
 rseness of the expanded representation. In addition\, the low dimensionali
 ty of the input layer generates overlaps between the induced representatio
 ns of different stimuli\, limiting the benefit of expansion. Highly sparse
  expansive representations obtained through synapses that encode the clust
 ered structure of the input reduce both intrastimulus variability and the 
 excess overlaps between stimuli\, enhancing the ability of downstream neur
 ons to perform classification and recognition tasks. Implications for olfa
 ctory\, cerebellar\, and visual processing are discussed.
LOCATION:Cambridge University Engineering Dept.\, Board Room (2nd floor\, 
 Baker building)
END:VEVENT
END:VCALENDAR