BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The Principle of Least Cognitive Action - Professor Marco Gori
DTSTART:20190529T151500Z
DTEND:20190529T160000Z
UID:TALK125104@talks.cam.ac.uk
CONTACT:jo de bono
DESCRIPTION:In this talk we introduce the principle of Least Cognitive Act
 ion with the purpose of understanding perceptual learning processes. The p
 rinciple closely parallels related approaches in physics\, and suggests to
  regard neural networks as systems whose weights are Lagrangian variables\
 , namely functions depending on time. Interestingly\, neural networks “c
 onquer their own life” and there is no neat distinction between learning
  and test\; their behavior is characterized by the stationarity of the cog
 nitive action\, an appropriate functional which contains a potential and a
  kinetic term. While the potential term is somewhat related to the loss fu
 nction used in supervised and unsupervised learning\, the kinetic term rep
 resents the energy connected with the velocity of weight change. Unlike tr
 aditional gradient descent\, the stationarity of the cognitive action yiel
 ds differential equations in the connection weights\, and gives rise to a 
 dissipative process which is needed to yield ordered configurations. We gi
 ve conditions under which this learning process reduces to stochastic grad
 ient descent and to Backpropagation. We give examples on supervised and un
 supervised learning\, and briefly discuss the application to deep convolut
 ional neural networks\, where an appropriate Lagrangian term is used to en
 force motion invariance in the visual feature extraction. \n\nVideo record
 ing available at:\nhttps://www.cl.cam.ac.uk/seminars/wednesday/video/lt2-1
 90529-wed-1600-t125104.html\n(note that the first 2:00 minutes of audio ar
 e missing).\n
LOCATION:Lecture Theatre 2\, Computer Laboratory
END:VEVENT
END:VCALENDAR