Statistics Seminar: Some Recent Results on Transfer Learning

Presenter: Oscar Hernan Madrid Padilla, Assistant Professor, University of California, Los Angeles

Description: In the first part of the talk, I will introduce TRansfer leArning via guideD horseshoE prioR (TRADER), a novel approach enabling multi-source transfer through pre-trained models in high-dimensional linear regression. TRADER shrinks target parameters towards a weighted average of source estimates, accommodating sources with different scales. Theoretical investigation shows that TRADER achieves faster posterior contraction rates than standard continuous shrinkage priors when sources align well with the target while preventing negative transfer from heterogeneous sources. Extensive numerical studies and a real-data application demonstrate that TRADER improves estimation and inference accuracy over state-of-the-art transfer learning methods. In the second part of the talk, I will discuss some ongoing work involving transfer learning in nonparametric regression with ReLU networks

Bio: Oscar Madrid Padilla is a tenure-track Assistant Professor in the Department of Statistics at the University of California, Los Angeles. Previously, from July 2017 to June 2019, he was a Neyman Visiting Assistant Professor in the Department of Statistics at the University of California, Berkeley. Before that, he earned his Ph.D. in Statistics from The University of Texas at Austin in May 2017 under the supervision of Professor James Scott. He completed his undergraduate degree, a B.S. in Mathematics, at CIMAT in Mexico in April 2013.

Hosted by: Statistics Department