JSM2023
Back to the program
Topic-Contributed Paper Session

New and Unexpected Risk Landscapes for Overparameterized Models in Machine Learning

Wed, Aug 9, 8:30 AM
Yoonkyung LeeChair
Section on Statistical Learning and Data Science

4 Presentations

Loss landscapes and optimization in overparameterized nonlinear systems and neural networks
8:35 AM - 8:55 AM
Mikhail Belkin
Co-authors: Chaoyue Liu (UCSD), Libin Zhu (UCSD)
How many parameters does your interpolator use? Revisiting degrees of freedom in the wake of overparametrized machine learning
8:55 AM - 9:15 AM
Ryan Tibshirani (Carnegie Mellon University)
Co-authors: Pratik Patil (University of California, Berkeley)
Predictive Model Degrees of Freedom in Linear Regression
9:15 AM - 9:35 AM
Yunzhang Zhu (The Ohio State University)
Co-authors: Yoonkyung Lee (The Ohio State University), Bo Luan (The Ohio State University)
Benefit of Interpolation in Nearest Neighbor Algorithms
9:35 AM - 9:55 AM
Yue Xing (Purdue University)
Co-authors: Qifan Song, Guang Cheng (Purdue University)
↑