BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//UM//UM*Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:America/Detroit
TZURL:http://tzurl.org/zoneinfo/America/Detroit
X-LIC-LOCATION:America/Detroit
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20070311T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20071104T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20261113T161030
DTSTART;TZID=America/Detroit:20261120T160000
DTEND;TZID=America/Detroit:20261120T170000
SUMMARY:Workshop / Seminar:Beyond Problem Solving: AI for Problem Generation\, Abstraction and Generalization
DESCRIPTION:AI has made remarkable progress in solving well-defined mathematical problems\, advancing in just a few years from grade-school exercises to helping settle problems at the research frontier. In this talk\, I'll present recent and ongoing work from my new lab at U-M CSE on AI systems for tasks beyond problem solving\, with applications to both formal mathematics and software. I'll first present The Token Games\, an evaluation framework in which AI models challenge each other with puzzles they generate\, inspired by Renaissance-era mathematical duels. We find that scores from The Token Games strongly correlate with performance on existing reasoning benchmarks\, while being cheaper to obtain and easier to extend\, since it's problem pool is not fixed. However\, while useful for evaluation\, the problems generated in these duels are largely \"uninteresting\" to humans. In another recent work\, we compared human and LLM judgments of problem interestingness\, finding that while some models agree substantially with humans on which problems are interesting\, they often fail to identify why humans find them interesting. Finally\, I'll briefly discuss preliminary work on learning useful\, reusable abstractions from formal proofs\, including generalizing theorems\, synthesizing proof tactics\, and discovering new definitions through symbolic program compression.\n\nBio: Gabriel Poesia is a new Assistant Professor of Computer Science and Engineering at the University of Michigan\, Ann Arbor. His research lies at the intersection of formal reasoning and AI\, grounded in applications in program verification and formal mathematics. He recently completed his PhD in Computer Science at Stanford University in 2025\, and spent a year as a Research Fellow at the Kempner Institute at Harvard University before starting at Michigan.
UID:153554-21916001@events.umich.edu
URL:https://events.umich.edu/event/153554
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:Artificial Intelligence,Mathematics
LOCATION:East Hall - 1360
CONTACT:
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20261120T162043
DTSTART;TZID=America/Detroit:20261120T163000
DTEND;TZID=America/Detroit:20261120T180000
SUMMARY:Conference / Symposium:Markley (2026-2027) (Housing)
DESCRIPTION:
UID:150757-21917982@events.umich.edu
URL:https://events.umich.edu/event/150757
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:Sessions
LOCATION:Angela Davis Lounge
CONTACT:
END:VEVENT
END:VCALENDAR