BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Cog Lunch: Guy Gaziv
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260517T065009Z
UID:tag:localist.com\,2008:EventInstance_48792522235802
DTSTART:20250211T170000Z
DTEND:20250211T180000Z
DESCRIPTION:Zoom link: https://mit.zoom.us/j/98651233056\n\nSpeaker: Guy Ga
 ziv\n\nAffiliation: DiCarlo Lab\n\nTitle: Robustified ANNs Reveal Wormhole
 s Between Human Category Percepts\n\nAbstract: \n\nThe visual object categ
 ory reports of artificial neural networks (ANNs) are notoriously sensitive
  to tiny\, adversarial image perturbations. Because human category reports
  (aka human percepts) are thought to be insensitive to those same small-no
 rm perturbations — and locally stable in general — this argues that AN
 Ns are incomplete scientific models of human visual perception. Consistent
  with this\, we show that when small-norm image perturbations are generate
 d by standard ANN models\, human object category percepts are indeed highl
 y stable. However\, in this very same "human-presumed-stable" regime\, we 
 find that robustified ANNs reliably discover low-norm image perturbations 
 that strongly disrupt human percepts. These previously undetectable human 
 perceptual disruptions are massive in amplitude\, approaching the same lev
 el of sensitivity seen in robustified ANNs. Further\, we show that robusti
 fied ANNs support precise perceptual state interventions: they guide the c
 onstruction of low-norm image perturbations that strongly alter human cate
 gory percepts toward specific prescribed percepts. These observations sugg
 est that for arbitrary starting points in image space\, there exists a set
  of nearby "wormholes"\, each leading the subject from their current categ
 ory perceptual state into a semantically very different state. Moreover\, 
 contemporary ANN models of biological visual processing are now accurate e
 nough to consistently guide us to those portals.
GEO:42.362302;-71.091766
LOCATION:Building 46\, 3310
SUMMARY:Cog Lunch: Guy Gaziv
URL;VALUE=URI:https://calendar.mit.edu/event/cog-lunch-guy-gaziv
END:VEVENT
END:VCALENDAR