BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Deep-layered machines have a built-in Occam's razor - Dr Thomas Fi
 nk\, Director\, London Institute for Mathematical Sciences
DTSTART:20260311T120000Z
DTEND:20260311T130000Z
UID:TALK245611@talks.cam.ac.uk
CONTACT:Challenger Mishra
DESCRIPTION:Input-output maps are prevalent throughout science and technol
 ogy. They are empirically observed to be biased towards simple outputs\, b
 ut we don't understand why. To address this puzzle\, we study the archetyp
 al input-output map: a deep-layered machine in which every node is a Boole
 an function of all the nodes below it. We give an exact theory for the dis
 tribution of outputs\, and we confirm our predictions through extensive co
 mputer experiments. As the network depth increases\, the distribution beco
 mes exponentially biased towards simple outputs. This suggests that deep-l
 ayered machines and other learning methodologies may be inherently biased 
 towards simplicity in the models that they generate.\n\nPreprint: https://
 arxiv.org/abs/2603.01217\n\nZoom: \nhttps://cl-cam-ac-uk.zoom.us/j/6590822
 098\nMeeting ID: 659 082 2098\nPasscode: 1dYRka\n
LOCATION:Computer Lab\, FW26 and Online (link in abstract)
END:VEVENT
END:VCALENDAR