BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Information Lattice Learning
DTSTART:20260910T101500
DTEND:20260910T111500
DTSTAMP:20260916T060549Z
UID:08828108ec9608187dcf31ea1a488a230dddc35fe49286d6ea58c684
CATEGORIES:Conferences - Seminars
DESCRIPTION:By: Lav Varshney - Stony Brook University\nVideo of his talk\
 n\n\nAbstract\nDrawing on group-theoretic and information-theoretic found
 ations that go back to Shannon's lattice theory of information\, we propos
 e information lattice learning (ILL) as a general framework to learn r
 ules of a signal. In our definition\, a rule is a coarsened signal used to
  help us gain interpretable insights about the original signal. To make fu
 ll sense of what might govern the signal’s intrinsic structure\, we see
 k multiple disentangled rules arranged in a hierarchy\, called a lattice.
   Compared to representation/rule-learning models optimized for a speci
 fic task (e.g.\, classification)\, ILL focuses on explainability: it is d
 esigned to mimic human experiential learning and discover rules akin to 
 those humans can distill and comprehend. We detail the mathematical foun
 dations and algorithms of ILL\, and illustrate how it addresses the funda
 mental question “what makes X an X” by creating rule-based explanatio
 ns designed to help humans understand. We show ILL’s efficacy and inter
 pretability on benchmarks and assessments in visual classification\, sign
 ificantly outperforming vision transformers. We further show optimality in
  semantic compression and the ability to create formally-verifiable digita
 l twins.  We close with applications in knowledge discovery and creativi
 ty\, using ILL to distill music theory from scores and enabling a new way
  to compose music\, as well as some early work on understanding the princ
 iples that govern scattering amplitudes in Super Yang-Mills theory\, rath
 er than just predicting them.\n\nBio\nLav R. Varshney is the Della Piet
 ra Infinity Professor and inaugural director of the AI Innovation Insti
 tute at Stony Brook University. He is co-founder and CEO of Kocree\, In
 c.\, a startup company building novel human-controllable AI for discover
 y and creativity\, and chief scientist of Ensaras\, Inc.\, a startup com
 pany focused on AI and wastewater treatment. He holds appointments at RAN
 D Corporation and at Brookhaven National Laboratory. He was previously on
  the faculty of the University of Illinois Urbana-Champaign\, a visiting
  scholar at Northwestern's Kellogg School of Management\, a principal rese
 arch scientist at Salesforce Research AI where he was part of the team th
 at developed and open weight released the first billion-parameter large l
 anguage model\, and a research staff member at IBM Research where he led
  the design and deployment of the first commercially-successful generativ
 e AI technology. He is a former White House staffer\, having served on
  the National Security Council staff as a White House Fellow\, where he
  contributed to national/international AI and wireless communications pol
 icy. His research interests include information theory and artificial in
 telligence. He received his B.S. degree from Cornell University and his S
 .M. and Ph.D. degrees from the Massachusetts Institute of Technology.  H
 e studied at EPFL in 2006 with Emre Telatar and Ruediger Urbanke.\n\nMore 
 information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/
 j/61827199715
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR