BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20250812T095123EDT-8060EmZWrg@132.216.98.100
DTSTAMP:20250812T135123Z
DESCRIPTION:Title: Top-down optimization recovers biological coding princip
 les of single-neuron adaptation in RNNs.\n\nAbstract: Spike frequency adap
 tation (SFA) is a well studied physiological mechanism with established co
 mputational properties at the single neuron level\, including noise mitiga
 ting effects based on efficient coding principles. Network models with ada
 ptive neurons have revealed advantages including modulation of total activ
 ity\, supporting Bayesian inference\, and allowing computations over distr
 ibuted timescales. Such efforts are bottom-up\, modeling adaptive mechanis
 ms from physiology and analysing their effects. How top-down environmental
  and functional pressures influence the specificity of adaptation remains 
 largely unexplored.\n	\n	In this talk\, I will discuss work where we use dee
 p learning to uncover optimal adaptation strategies from scratch\, in recu
 rrent neural networks (RNNs) performing perceptual tasks. In our RNN model
 \, each neuron's activation function (AF) is taken from a parametrized fam
 ily to allow modulation mimicking SFA\, and an adaptation controller is tr
 ained end-to-end to control an AF in real time\, based on pre-activation i
 nputs to a neuron. Remarkably\, we find emergent adaptation strategies tha
 t implement SFA mechanisms from biological neurons\, including fractional 
 input differentiation. This suggests that even in simplified models\, envi
 ronmental pressures and objective-based optimization are enough for sophis
 ticated biological mechanisms to emerge.\n\n \n\nCAMBAM Seminar Series\n	En
  ligne/Web - https://mcgill.zoom.us/j/85428056343\n
DTSTART:20211214T170000Z
DTEND:20211214T180000Z
SUMMARY:Guillaume Lajoie (University of Montreal)
URL:/mathstat/channels/event/guillaume-lajoie-universi
 ty-montreal-335392
END:VEVENT
END:VCALENDAR