BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Prospective coding by spiking neurons - Johanni Brea (EPFL)
DTSTART:20161115T090000Z
DTEND:20161115T100000Z
UID:TALK69176@talks.cam.ac.uk
CONTACT:Rodrigo Echeveste
DESCRIPTION:Brains can learn to predict. But can a single neuron do so?  B
 uilding on the work of Urbanczik & Senn (2014) on learning by the dendriti
 c prediction of somatic spiking\, a plasticity rule that implements superv
 ised\, unsupervised and reinforcement learning in a spiking neuron model\,
  I will show that a slightly longer window of synaptic potentiation allows
  a spiking neuron to match its current firing rate to its own expected fut
 ure discounted firing rate.  For instance\, if an originally neutral event
  is repeatedly followed by an event that\nelevates the firing rate of a ne
 uron\, the originally neutral event will eventually also elevate the neuro
 n's firing rate. The plasticity rule is a form of spike timing dependent p
 lasticity in which a presynaptic spike followed by a postsynaptic spike le
 ads to potentiation. Even if the plasticity window has a width of 20 milli
 seconds\, associations on the time scale of seconds can be learned. I will
  illustrate prospective coding with three examples: learning to predict a 
 time varying input\, learning to predict the next stimulus in a delayed pa
 ired-associate task and learning with a recurrent network to reproduce a t
 emporally compressed version of a sequence. In the special case that the s
 ignal to be predicted encodes reward\, the neuron learns to predict the di
 scounted future reward and learning is closely related to the temporal dif
 ference learning algorithm TD(lambda).
LOCATION:Cambridge University Engineering Department\, CBL\, BE-438 (http:
 //learning.eng.cam.ac.uk/Public/Directions)
END:VEVENT
END:VCALENDAR