BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Learning to Unscramble with Agentic AI: A New Self-Supervised Appr
 oach to Symbolic Simplification
DTSTART:20260612T150000Z
DTEND:20260612T170000Z
DTSTAMP:20260604T080900Z
UID:indico-event-33127@indico.bnl.gov
DESCRIPTION:Speakers: David Shih (Rutgers)\n\nAbstract: I will give an in-
 depth overview of my two recent papers [2603.11164 and 2604.05034] on a hi
 ghly effective new ML method for symbolic simplification of mathematical e
 xpressions based on a Markov Decision Process trained with self-supervised
  learning. Training data is generated automatically by scrambling simple e
 xpressions into complicated ones using a set of mathematical identities\, 
 and recording the reverse unscrambling steps. When applied to dilog sums a
 nd tree-level YM scattering amplitudes\, our method is shown to have nearl
 y perfect simplification rate\, far surpassing previous approaches based o
 n reinforcement learning and end-to-end regression. When extended to IBP r
 eduction\, we show (in the toy 3pt two-loop triangle-box setting) that the
  method is able to reduce integrals of increasing weight while remaining f
 lat in memory consumption\, compared to general purpose Laporta-based algo
 rithms such as Kira where the memory consumption rises rapidly with weight
 . These projects were both done in collaboration with Claude Code\, and I 
 will conclude with some lessons learned and general thoughts for the futur
 e of AI and our field.\n\nhttps://indico.bnl.gov/event/33127/
LOCATION:CFNS Library (https://bnl.zoomgov.com/j/1614715193?pwd=WkwxODVWdz
 Zzb29zQnZRVGp3VTBDQT09)
URL:https://indico.bnl.gov/event/33127/
END:VEVENT
END:VCALENDAR