An Occam’s Razor Journey in Deep Learning: Drawing Inspiration from Applications

Title: An Occam’s Razor Journey in Deep Learning: Drawing Inspiration from Applications

Abstract: The enthusiasm for deep learning has sparked a plethora of neural network models and learning algorithms. However, when it comes to particular applications, such a plethora makes it hard to decide what model or algorithm to use—there are simply too many options! Quite often, the SOTA algorithms on the benchmark leaderboards are not necessarily the best. One may blame particular applications have too specific needs, but from a different viewpoint, these needs offer valuable priors for us to use the old-fashioned Occam’s Razor principle to design and select algorithms. 

In this talk, I will present several works along my journey of using Occam’s Razor principle in deep learning research, drawing inspiration from applications in biology and self-driving. In particular, I will present our ongoing/recent works on fine-grained segmentation for specimen images, interpretable models for trait identification and discovery, building foundation models for the Tree of Life, and pre-training for 3D object detection. I will conclude this talk with quick overviews of other examples where Occam’s Razor principle works in deep learning.

Bio: Wei-Lun (Harry) Chao (https://cse.osu.edu/people/chao.209) is a Distinguished Assistant Professor in the Department of Computer Science and Engineering (CSE) at the Ohio State University (OSU). His research interests are machine learning and computer vision and their applications in visual recognition, autonomous driving, biology, and healthcare, seeking to establish fundamental understandings and develop robust and widely applicable algorithms to resolve real-world challenges. He is particularly interested in learning from imperfect data problems in these domains, including limited, noise, heterogeneous, distribution-shifting, and inaccessible data. He was awarded the OSU College of Engineering Lumley Research Award in 2023, the OSU CSE Faculty Teaching Award in 2024, and a co-author of the CVPR 2024 Best Student Paper Award. 

Before joining OSU, he was a Postdoctoral Associate at Cornell University, working with Kilian Weinberger, Mark Campbell, and Bharath Hariharan. He received his Ph.D. in Computer Science from the University of Southern California under the supervision of Fei Sha.