Academy of Mathematics and Systems Science, CAS
Colloquia & Seminars

Speaker:	Professor Yik-Chung Wu,University of Hong Kong
Inviter:	Associate Professor Yafeng Liu
Title:	Tunning-free matrix and tensor factorizations for signal processing and machine learning
Language:	Chinese
Time & Venue:	2024.03.25 10:30-11:30 S515
Abstract:	Matrix and Tensor factorizations are important data analytic tools in many applications, such as recommendation systems, image completion, social network data mining, wireless communications, etc. Traditionally, matrix and tensor factorizations are approached from optimization perspective. While proven to be effective, optimization-based matrix and tensor factorizations usually involve hyperparameters tuning, with one of the major hyperparameters being the matrix or tensor rank. However, when the number of hyperparameters is more than 3 or 4, tuning them becomes computationally expensive. This talk approaches the problem from the Bayesian perspective and shows how hyperparameter tuning can be eliminated while providing comparable or even better performance than corresponding optimization-based algorithms.