CAN AI MODELS BE LIKE SCIENTIFIC INSTRUMENTS?
Information Theory Society Distinguished Lecture
Modern ML/AI models are incredibly complex. Their size makes them expensive to train and use. Sometimes AI systems are touted as the new "scientific instruments" but what does that mean? In this talk I will discuss recent work which tries to get at some basic questions: how should we measure how models differ from each other and how do they change through training? We can use some tools from "old school" statistics/probability to get some handle on this in terms of understanding variability during training and using AI models as "instruments" to look at other models. While much of this work is empirical, the findings point to some interesting directions for theory and engineering.
This talk is based on joint work with Sinjini Banerjee, Reilly Cannon, Sutenay Choudhury, Tony Chiang, Ioana Dumitriu, Andrew Engel, Natalie Frank, Tim Marrinan, Max Vargas, and Zhichao Wang.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
- Ian Polmear Meeting Room
- 14 Alliance Lane, Monash University
- Clayton, Victoria
- Australia
- Building: 72
- Room Number: 210
- Contact Event Host
- Co-sponsored by The University of Melbourne
- Starts 05 August 2025 02:00 PM UTC
- Ends 15 August 2025 12:01 AM UTC
- No Admission Charge
Speakers
Anand Sarwate of Rutgers University, USA
CAN AI MODELS BE LIKE SCIENTIFIC INSTRUMENTS?
Modern ML/AI models are incredibly complex. Their size makes them expensive to train and use. Sometimes AI systems are touted as the new "scientific instruments" but what does that mean? In this talk I will discuss recent work which tries to get at some basic questions: how should we measure how models differ from each other and how do they change through training? We can use some tools from "old school" statistics/probability to get some handle on this in terms of understanding variability during training and using AI models as "instruments" to look at other models. While much of this work is empirical, the findings point to some interesting directions for theory and engineering.
This talk is based on joint work with Sinjini Banerjee, Reilly Cannon, Sutenay Choudhury, Tony Chiang, Ioana Dumitriu, Andrew Engel, Natalie Frank, Tim Marrinan, Max Vargas, and Zhichao Wang.
Biography:
Anand D. Sarwate is a professor in the Electrical and Computer Engineering Department at Rutgers, The State University of New Jersey. He received B.S. degrees in mathematics and electrical engineering from MIT and a Ph.D. in electrical engineering from UC Berkeley. Prior to joining Rutgers he was a Research Assistant Professor at TTI-Chicago and a postdoc at the ITA Center at UC San Diego. His research interests include information theory, machine learning, signal processing, optimization, and privacy and security. Dr. Sarwate serves on the Board of Governors of the IEEE Information Theory Society (ITSOC) and is a ITSOC Distinguished Lecturer for 2024-2025.
Address:Rutgers University, , New Jersey, United States