CSE Faculty Candidate Seminar - Jiayun (Peter) Wang

Name: Jiayun (Peter) Wang, Postdoctoral Researcher from Caltech

Date: Tuesday, April 1, 2025 at 11:00 am

Location: Scheller College of Business, Room 202 (Google Maps link)

Link: The recording of this in-person seminar will be uploaded to CSE's MediaSpace

Coffee and snacks provided!

Title: Any-Resolution Neural Operator for Inverse Problems

Abstract: Inverse problems, the process of estimating the latent cause that explains the observations, are some of the most important problems in science. Solving inverse problems like computational imaging is fundamental to many fields—from healthcare diagnostics to decision-making—yet the problem is often challenging because observations can be incomplete or low-resolution. Traditional model-based methods struggle with these ill-posed inverse problems, where multiple solutions can explain the same data. Neural networks, another way to solve the inverse problem, can overfit to specific resolutions. To tackle the challenges, this talk introduces operator learning that approximates solution operators for inverse problems in any resolution. By incorporating physics-based loss functions (such as enforcing known equations that model the forward problem of photoacoustic imaging) and designing resolution-agnostic architectures (via signal decomposition on bases) that adapt and learn seamlessly to different resolutions (such as MRI and ultrasound imaging), our approach reconstructs high-quality images from real-world observations. My future work will explore eliminating the need for paired data for training neural operators and learning emergent features via self-supervised learning, supported by my previous work on self-supervised representation learning that leverages spatial and temporal co-occurrence to capture data similarity, with applications in ocular disease diagnosis and assessing surgeon skills from unlabeled videos. I will also explore integrating sensing with feature learning and combining forward and inverse problem formulations to broaden their application across science and engineering.

Bio: Jiayun (Peter) Wang is a postdoctoral researcher at the California Institute of Technology, working with Prof. Anima Anandkumar. He received his Ph.D. from UC Berkeley in 2023, advised by Prof. Stella Yu. He develops core machine learning and computer vision methods for healthcare applications. His work has been recognized with three Best Paper Awards (Machine Learning for Health 2023 & 2024, CVPR Workshop 2019). His research is supported by the NIH, NSF, ONR, and private foundations including the Roberta Smith Fund and Schmidt Sciences. He also gained industry experience at Amazon and Aizip, where he focused on translating research into robust and efficient on-device AI solutions. More information can be found at his website: https://pwang.pw/.