BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Which data are learnable?
DTSTART:20240416T121500
DTEND:20240416T140000
DTSTAMP:20260403T224647Z
UID:ace6bcee0e1bacaad817a9519ceec8c159d458eb1e6266880503310e
CATEGORIES:Conferences - Seminars
DESCRIPTION:Matthieu Wyart\nLearning generic tasks in high dimension is im
 possible\, as it would require an unreachable number of training data. Yet
  algorithms or humans can play the game of go\, decide what is on an image
  or learn languages. \nThe only resolution of this paradox is that learna
 ble data are highly structured. I will review ideas in the field of what t
 his structure may be. I will then discuss two recent results of our group.
  (i) if the data  is hierarchical\,\nsupervised learning can occur with a
  training set size which is polynomial\, and not exponential\, in the data
  dimension (1). (ii) I will discuss how novel data\, such as the pictures 
 of the fake celebrities below\, can be generated by composing\nknown featu
 res into a new whole. This theory of composition predicts a phase transiti
 on in generative models. I will discuss empirical evidence for its validit
 y (2). \n\n(1) Cagnetta\, Petrini\, Tomasini\, Favero\, Wyart\,  23’\n
 (2) Sclocchi\, Favero\, Wyart\, 24'\n\n\n\n\n\n 
LOCATION:BSP 727 https://plan.epfl.ch/?room==BSP%20727
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR