BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Testing AI's Implicit World Models
DTSTART:20260302T101500
DTEND:20260302T111500
DTSTAMP:20260526T115307Z
UID:8502fc09e949c5746060dc42bfaa8ff3be02b581040be2116bd80434
CATEGORIES:Conferences - Seminars
DESCRIPTION:Par : Keyon Vafa - Harvard University\nIC Faculty candidate\n\
 nAbstract\nReal-world AI systems must be robust across a wide range of con
 ditions. One path to such robustness is if a model recovers a coherent str
 uctural understanding of its domain. But it is unclear how to measure\, or
  even define\, structural understanding. This talk will present theoretica
 lly-grounded definitions and metrics that test the structural recovery —
  or implicit “world models” — of generative models. We will propose 
 different ways to formalize the concept of a world model\, develop tests b
 ased on these notions\, and apply them across domains. In applications ran
 ging from testing whether LLMs apply logic to whether foundation models ac
 quire Newtonian mechanics\, we will see that models can make highly accura
 te predictions with incoherent world models. We will also connect these te
 sts to a broader agenda of building generative models that are robust acro
 ss downstream uses\, incorporating ideas from statistics and the behaviora
 l sciences. Developing reliable inferences about model behavior across tas
 ks offer new ways to assess and improve the efficacy of generative models.
 \n\nBio\nKeyon Vafa is a postdoctoral fellow at Harvard University and an 
 affiliate with the Laboratory for Information & Decision Systems at MIT. H
 is research focuses on understanding and improving the implicit world mode
 ls learned by generative models. Keyon completed his PhD in computer scien
 ce from Columbia University\, where he was an NSF GRFP Fellow and the reci
 pient of the Morton B. Friedman Memorial Prize for excellence in engineeri
 ng. He has organized the NeurIPS 2024 Workshop on Behavioral Machine Learn
 ing and the ICML 2025 Workshop on Assessing World Models\, and serves on t
 he Early Career Board of the Harvard Data Science Review.\n\nMore informat
 ion\n 
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/
 j/62765156059
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR