BIU Learning Club, December 1, 2024 (Note, Sunday): Generalization in Overparameterized Machine Learning
December 1 @ 12:00 pm - 1:00 pm IST
On December 1 (Note this is Sunday), Dr. Yehuda Dar from the Ben Gurion University will give a talk titled: Generalization in Overparameterized Machine Learning
Abstract:
Modern machine learning models are highly overparameterized (i.e., they are very complex with many more parameters than the number of training data examples); yet, these models often generalize extremely well to inputs outside of their training dataset. This practical generalization performance motivates numerous foundational research questions that fall outside the scope of conventional machine learning concepts.
This talk will focus on statistical theory for transfer learning between source and target linear regression problems that are related and overparameterized. The theory shows how the generalization error of the target task is influenced by the transfer learning aspects, and characterizes the cases where transfer learning is beneficial. Specifically, the transfer learning theory elucidates the double descent phenomenon, which is an overparameterized generalization behavior that extends the classical bias-variance tradeoff. Moreover, empirical insights on the generalization and double descent phenomenon in transfer learning of deep neural networks will be presented.
BIO:
Yehuda Dar is an Assistant Professor in the Computer Science Department at Ben-Gurion University. His research is mainly on the theoretical and empirical foundations of deep learning and overparameterized machine learning. Yehuda holds a BSc in Computer Engineering, MSc in Electrical Engineering, and a PhD in Computer Science, all from the Technion — Israel Institute of Technology. Later on, he had a postdoctoral research position in the Department of Electrical and Computer Engineering at Rice University.