BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The Pervasive Role of Composing Transformations in Machine Learnin
 g - Anders Karlsson (Université de Genève)
DTSTART:20250722T135000Z
DTEND:20250722T144000Z
UID:TALK234607@talks.cam.ac.uk
DESCRIPTION:From the layer maps of neural networks to training procedures 
 and reinforcement learning\, compositions of transformations permeate mode
 rn AI. These compositional products often involve randomly selected maps\,
  as in weight initialization\, stochastic gradient descent (SGD)\, and dro
 pout. In reinforcement learning\, Bellman-type operators with randomness a
 re iterated to update reward structures and strategies. I will discuss the
  mathematics and geometry underlying the composition of random transformat
 ions. In particular\, I will explain a general limit law established in jo
 int work with Gou&euml\;zel. Moreover\, I will discuss a possible cut-off 
 phenomenon related to the depth of neural networks and the influence of it
 eration order. Motivated by these observations\, and in collaboration with
  Avelin\, Dherin\, Gonzalvo\, Mazzawi\, and Munn\, we propose backward var
 iants of SGD that improve stability and convergence while maintaining gene
 ralisation.&nbsp\;
LOCATION:External
END:VEVENT
END:VCALENDAR