Speaker
Xuefeng Du is a CS Ph.D. student at UW-Madison. His research focus is trustworthy machine learning, such as adversarial robustness, learning problem with label noise, out-of-distribution detection. He is also interested in neural architecture search, graph mining and high-level recognition models, such as object detection and segmentation.
Abstract
Out-of-distribution (OOD) detection is indispensable for deploying machine learning models in the wild. One of the key challenges in OOD detection is that models lack supervision signals from unknown data. Previous approaches rely on real outlier datasets for model regularization, which can be costly and sometimes infeasible to obtain in practice. In this talk, we are going to talk about our recent approaches to unifying representation learning and outlier synthesis to generate unknowns in a convenient and effective way. Then, we design sophisticated regularization terms to help our model achieve strong OOD detection performance for both image-level and object-level object detection.