BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Computational Neuroscience Journal Club - Dhruva Raman (University
  of Cambridge)
DTSTART:20190611T150000Z
DTEND:20190611T160000Z
UID:TALK126190@talks.cam.ac.uk
CONTACT:Rodrigo Echeveste
DESCRIPTION:Dhruva Raman will present:\n\n• Cerebellar learning using pe
 rturbations\n\n• Guy Bouvier\, Johnatan Aljadeff\, Claudia Clopath\, Cé
 lian Bimbard\, Jonas Ranft\, Antonin Blot\, Jean-Pierre Nadal\, Nicolas Br
 unel\, Vincent Hakim\, Boris Barbour\n\n• eLIFE (2018)\n\n• https://cd
 n.elifesciences.org/articles/31599/elife-31599-v1.pdf\n\nAbstract: The cer
 ebellum aids the learning of fast\, coordinated movements. According to\nc
 urrent consensus\, erroneously active parallel fibre synapses are depresse
 d by complex spikes\nsignalling movement errors. However\, this theory can
 not solve the credit assignment problem of\nprocessing a global movement e
 valuation into multiple cell-specific error signals. We identify a\npossib
 le implementation of an algorithm solving this problem\, whereby spontaneo
 us complex\nspikes perturb ongoing movements\, create eligibility traces a
 nd signal error changes guiding\nplasticity. Error changes are extracted b
 y adaptively cancelling the average error. This framework\,\nstochastic gr
 adient descent with estimated global errors (SGDEGE)\, predicts synaptic p
 lasticity\nrules that apparently contradict the current consensus but were
  supported by plasticity\nexperiments in slices from mice under conditions
  designed to be physiological\, highlighting the\nsensitivity of plasticit
 y studies to experimental conditions. We analyse the algorithm’s converg
 ence\nand capacity. Finally\, we suggest SGDEGE may also operate in the ba
 sal ganglia.
LOCATION:Cambridge University Engineering Department\, CBL\, BE4-38 (http:
 //learning.eng.cam.ac.uk/Public/Directions)
END:VEVENT
END:VCALENDAR