BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:Europe/Vienna
BEGIN:DAYLIGHT
DTSTART:20260329T030000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20261025T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260820T080148Z
UID:6a85953fddda5101667480@ist.ac.at
DTSTART:20260824T140000
DTEND:20260824T150000
DESCRIPTION:Speaker: Farzan Farnia\nhosted by Marco Mondelli\nAbstract: Dee
 p generative models have achieved remarkable success by learning to imitat
 e complex data distributions\, yet distributional fidelity alone does not 
 explicitly promote properties such as diversity\, novelty\, or imagination
 . In this talk\, I will explore how entropy can provide a principled way t
 o extend the conventional objective of generative modeling beyond imitatio
 n. Utilizing pretrained neural network representations\, we introduce a sp
 ectral notion of entropy that captures semantic diversity through the eige
 nvalues of a kernel covariance operator. I will discuss that the concavity
  of this entropy has two important consequences. First\, finite samples sy
 stematically underestimate the entropy of the underlying population\, moti
 vating our study of whether this diversity loss persists in trained genera
 tive models. Our experiments then reveal a consistent entropy gap between 
 standard generative models and their training distributions. Second\, conc
 avity makes entropy superlevel sets convex\, allowing us to show that proj
 ecting a low-diversity model onto a sufficiently high-entropy set can move
  it closer to the true data distribution. Building on these insights\, I w
 ill introduce Imaginative Generative AI (IGA)\, an entropy-regularized fra
 mework that provides a path from diversity correction to deliberate extrap
 olation\, with the entropy of the data defining an Entropy Wall that separ
 ates faithful imitation from the generation of more diverse worlds beyond 
 the observed distribution.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 )\, ISTA
ORGANIZER:swiddman@ist.ac.at
SUMMARY:Farzan Farnia: Entropy in Generative AI Models: From Imitation to I
 magination
URL:https://talks-calendar.ista.ac.at/events/6592
END:VEVENT
END:VCALENDAR