BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//6.4.7.2//EN
TZID:Europe/Paris
X-WR-TIMEZONE:Europe/Paris
BEGIN:VEVENT
UID:0-998@lptms.universite-paris-saclay.fr
DTSTART;TZID=Europe/Paris:20241112T110000
DTEND;TZID=Europe/Paris:20241112T120000
DTSTAMP:20241111T171511Z
URL:http://www.lptms.universite-paris-saclay.fr/seminars/seminaire-du-lptm
 s-marcelo-guzman-upenn/
SUMMARY:Séminaire du LPTMS : Marcelo Guzmán (UPenn) - Salle des séminair
 es du FAST et du LPTMS\, bâtiment Pascal n°530 - 12 Nov 24 11:00
DESCRIPTION:Learning functionality under physical constraints: how physics 
 shapes the way machines learn\n&nbsp\;\nMarcelo Guzmán (University of Pen
 nsylvania)\n&nbsp\;\n\nFrom biological systems to neuromorphic computing\,
  learning is fundamentally constrained by physics. These constraints\, r
 anging from optimization principles (e.g.\, energy minimization) to conser
 vation laws and stochastic dynamics in the presence of noise\, shape learn
 ing dynamics and learned functions in ways absent in artificial neural net
 works (ANNs). In this two-part talk\, I explore how physical constraints i
 nfluence learning by examining two paradigmatic physical learning models: 
 tunable mechanical networks and self-learning resistor networks.\n\nFirst\
 , I will show that learning in these physical networks is a dual optimizat
 ion problem. In the case of resistor networks\, for example\, it is the mi
 nimization of a cost with respect to conductances and the minimization of 
 the power dissipated with respect to voltages—the physical constraint. T
 his additional minimization couples cost and power\, enabling inference of
  key network components through simple physical measurements. I will demon
 strate that the high-curvature directions around the cost minima —highli
 ghting the key functional components— are captured by the network’s ph
 ysical susceptibilities. These susceptibilities\, encoded in the softest m
 odes of the power\, are measurable and provide clear insights into the net
 work’s functionality\, suggesting an interpretability advantage over ANN
 s and a new framework for studying biological systems for which the cost i
 s unknown.\n\nNext\, I will focus on the local dynamics of self-learning r
 esistor networks in laboratory settings. Learning in these systems is the 
 outcome of the collective behavior of individual components. While these n
 etworks are energy-efficient\, they are sensitive to the presence of exter
 nal noise and internal biases\, two physical constraints. I will show how 
 noise and bias affect the learning dynamics when training for two periodic
 ally alternating tasks. In ideal conditions\, periodic training converges 
 to an optimal solution for both tasks. However\, in the presence of noise 
 and bias\, learning leads to limit cycles in the space of conductances. Ba
 sed on theory and experiments\, I uncover a complex interplay between the 
 geometry of the solution space (linked to task complexity)\, bias\, and no
 ise\, revealing distinct learning phases in terms of the training period. 
 Finally\, I will show that under certain conditions\, bias can improve the
  networks’ learning capabilities.
CATEGORIES:seminars
LOCATION:Salle des séminaires du FAST et du LPTMS\, bâtiment Pascal n°53
 0\, rue André Riviere\, Orsay\, 91405\, France
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=rue André Riviere\, Orsay\
 , 91405\, France;X-APPLE-RADIUS=100;X-TITLE=Salle des séminaires du FAST 
 et du LPTMS\, bâtiment Pascal n°530:geo:0,0
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:STANDARD
DTSTART:20241027T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
END:VTIMEZONE
END:VCALENDAR