第65回統計的機械学習セミナー / The 65th Statistical Machine Learning Seminar (Hybrid)

【Date & Time】
May 30th (Friday), 2025　16:00 - 17:30
Admission Free
【Place】
Seminar Room 5 (D313/D314), ISM
Hybrid Registration: https://forms.gle/64XfC3pPs3jN2b5h9
(Please directly drop by at the seminar room if you join in person)
【Speaker】
Prof. Yutong Wang (Illinois Tech, USA)
【Title】
Reinventing the Foundations of Multiclass Classification
【Abstract】
Classification, a central task in machine learning, has evolved from recognizing handwritten characters with tens of classes to next-word prediction in language models with hundreds of thousands of classes. However, our theoretical understanding of classification has not kept pace with the rapid growth in task complexity. To address this, the first part of the talk will introduce a new classification framework that unifies binary and multiclass regimes. I will discuss how this framework is used to understand multiclass implicit regularization. I will also discuss how the framework reveals a surprising connection between multiclass SVM, ranking with ties allowed, and label noise.
【Speaker Bio】
Yutong Wang is an assistant professor at Illinois Tech starting Fall 2024. His research spans from theory of machine learning to AI for science. Previously, he was a Postdoctoral Research Fellow supported by the Eric and Wendy Schmidt AI Fellowship in the Department of Electrical and Computer Engineering (ECE) at the University of Michigan, Ann Arbor, collaborating with Qing Qu and Wei Hu. Before the postdoc, he completed PhD in ECE advised by Clay Scott, also at the University of Michigan.