BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Talk Cancelled: Less is more - Anita Faul
DTSTART:20241031T130000Z
DTEND:20241031T140000Z
UID:TALK219988@talks.cam.ac.uk
CONTACT:Jack Atkinson
DESCRIPTION:Please note that the speaker is no longer available on this da
 te so the talk has been cancelled. We will look at arranging an alternativ
 e.\n\nOccam's razor states "Plurality is never to be posited without neces
 sity." We begin by examining how small a neural network can distinguish th
 e digits in the MNIST data set. For continuous problems\, there are Univer
 sal Approximation Theorems. For any function and a criterion of closeness\
 , if there are enough neurons in a neural network\, then there exists a ne
 ural network with that many neurons that does approximate the function tha
 t close. However\, is this desirable? Simpler systems facilitate human ins
 ight. We look at the following challenges in data approximation also known
  as inference in machine learning.\n\n* Ill-conditioned matrices.\n* Model
 ing data\, keeping the model simple while explaining the data adequately.\
 n* Choice of model space.\n* New data arriving.\n* Model updates.\n* Confi
 dence in model predictions.\n* Informed data collection.\n* Limited comput
 ing power\, up and down-link capacity.\n\nAnd introduce a system which bui
 lds up complexity when it is necessary.\n\nThe zoom link is https://cam-ac
 -uk.zoom.us/j/81161988457?pwd=TB5DgLyL0RLQROGBA4LC9jLnlKAh5p.1 (Password 3
 55996)
LOCATION:Rayleigh Seminar Room\, Maxwell Centre\, West Cambridge
END:VEVENT
END:VCALENDAR