BEGIN:VCALENDAR
PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3-
 M3//EN
VERSION:2.0
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20251023T201414Z
DTSTART:20251103T140000Z
DTEND:20251103T150000Z
SUMMARY:Yasser Roudi -- Long-term behaviour of learning machines under cl
 osed-loop learning [IN PERSON]
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}dre-mfuztj2
 q-luas0q
DESCRIPTION:Join us for this seminar by Yasser Roudi (King's) as part of 
 the Maths in the Life Sciences seminar series (and the online North West
  Seminar Series in Mathematical Biology and Data Sciences in collaborati
 on with Liverpool Universities). \n\nTitle: Long-term behaviour of learn
 ing machines under closed-loop learning\n\nAbstract: In the standard set
 up for learning\, a learning machine\, whether a neural newtork\, statis
 tical model\, biological agent\, attempts to lear structure from data ge
 nerated from the environment. In a closed-loop learning\, on the other h
 and\, model parameters are repeatedly estimated from data generated from
  the model itself\, another model with a similar structure. Given the po
 ssibility that large neural network models may\, in the future\, be prim
 arily trained with data generated by artificial neural networks themselv
 es\, understanding the long-term behaviour of learning machines under cl
 osed-loop learning is receiving a lot of attention. \n\nIn this talk\, w
 e study this process in the case of models that belong to exponential fa
 milies. We show that in this case\, maximum likelihood estimation of the
  parameters results in a process that converges to absorbing states that
  amplify initial biases present in the data. This outcome may\, however\
 , be prevented if the data contains at least one data point generated fr
 om a ground truth model\, by relying on maximum a posteriori estimation 
 or by introducing regularisation. Simulations also show that similar res
 ults are observed in the Hopfield or Restricted Boltzmann Machines. \n\n
 Ref: https://arxiv.org/abs/2506.20623 [arxiv.org]\n\nThe talk will be al
 so be streamed via Teams\, please contact carl.whitfield@manchester.ac.u
 k or igor.chernyavsky@manchester.ac.uk for the zoom link\, or sign up to
  the mailing list.\n\nTo subscribe to the mailing list for this event se
 ries\, please send an e-mail with the phrase “subscribe math-lifesci-sem
 inar” in the message body to listserv@listserv.manchester.ac.uk
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:1.009\, Roscoe Building\, Manchester
END:VEVENT
END:VCALENDAR