BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20250802T164739EDT-4225uvt9BR@132.216.98.100
DTSTAMP:20250802T204739Z
DESCRIPTION:Top-down optimization recovers biological coding principles of 
 single-neuron adaptation in RNNs\n\nGuillaume Lajoie\, University of Montr
 eal\n	Tuesday December 14\, 12-1pm\n	Zoom Link: https://mcgill.zoom.us/j/854
 28056343\n\nAbstract: Spike frequency adaptation (SFA) is a well studied p
 hysiological mechanism with established computational properties at the si
 ngle neuron level\, including noise mitigating effects based on efficient 
 coding principles. Network models with adaptive neurons have revealed adva
 ntages including modulation of total activity\, supporting Bayesian infere
 nce\, and allowing computations over distributed timescales. Such efforts 
 are bottom-up\, modeling adaptive mechanisms from physiology and analysing
  their effects. How top-down environmental and functional pressures influe
 nce the specificity of adaptation remains largely unexplored.\n\nIn this t
 alk\, I will discuss work where we use deep learning to uncover optimal ad
 aptation strategies from scratch\, in recurrent neural networks (RNNs) per
 forming perceptual tasks. In our RNN model\, each neuron's activation func
 tion (AF) is taken from a parametrized family to allow modulation mimickin
 g SFA\, and an adaptation controller is trained end-to-end to control an A
 F in real time\, based on pre-activation inputs to a neuron. Remarkably\, 
 we find emergent adaptation strategies that implement SFA mechanisms from 
 biological neurons\, including fractional input differentiation. This sugg
 ests that even in simplified models\, environmental pressures and objectiv
 e-based optimization are enough for sophisticated biological mechanisms to
  emerge.\n
DTSTART:20211214T170000Z
DTEND:20211214T180000Z
LOCATION:CA\, QC
SUMMARY:QLS Seminar Series - Guillaume Lajoie
URL:/qls/channels/event/qls-seminar-series-guillaume-l
 ajoie-335095
END:VEVENT
END:VCALENDAR