Rishi Sonthalia
Assistant Professor
Mathematics, Boston College
rishi.sonthalia@bc.edu
Assistant Professor
Mathematics, Boston College
rishi.sonthalia@bc.edu
September - December 2024 - Los Angeles for IPAM program on Mathematics of Intelligence
October 9, 2024 - Yale for a seminar
October 21 - October 25, 2024- Atlanta for SIAM conference on Mathematics of Data Science
November 4, 2024 - Tufts for a seminar
November 14 - November 16, 2024 - Philadelphia for DeepMath
December 11 - December 15, 2024 - Vancouver for NeurIPS
December 16, 2024 - London for the session on “Over-parameterization in ML” at the CFE-CMStatistics conference
January 2025 - Seattle for JMM, organizing a session on Geometry and Combinatorics for Machine Learning with Eli Grigsby and Kathryn Lindsey.
January 2025 - Ethereum Foundations Eval Science Primer Series
March 2025 - Boston College, Computer Science Seminar
April 2025 - UCLA, Math Colloquium
July 2025 - ICML Workshop on High Dimensional Learning Dynamics
August 2025 - SIAM Annual Meeting, sessions on Double Descent
September 2025 - University of Massachussets Lowel Seminar
September 2025 - Hamburg for Conference on Mathematics of Machine Learning
Novermber 2025 - Univerisity of Pittsburgh, Workshop on Mathematical Analysis and Machine Learning.
March 2026 - Organizing MFO workshop on Novel and Modern Phenomena in Machine Learning
I am broadly interested in math for machine learning. I think math for machine learning can be used in one of two ways.
We use math to analyze existing methods and prove theoretical results. These results help us gain insights into techniques that are currently being used.
In this area, I am primarily interested in how regularization affects generalization. This would include both explicit regularization, such as ridge regularization, early stopping, and regularization via noise, as well as implicit regularization from the parameterization, optimization and the loss landscape.
I am also interested in other questions relating to universal approximation.
We can use math to design new principled methods.
I am interested in designing new geometric deep learning techniques using ideas from differential geometry and coarse geometric group theory.
I am also interested in generative AI, and using ideas from geometry and random matrix theory to improve their design.