Introduction to Professor Wei Hu and Her Lab at Peking University

1. Could you briefly introduce yourself (and your University/Lab)?
I am currently an Assistant Professor with Wangxuan Institute of Computer Technology, Peking University. I received the B.S. degree in Electrical Engineering from the University of Science and Technology of China in 2010, and the Ph.D. degree in Electronic and Computer Engineering from the Hong Kong University of Science and Technology in 2015. I was a Researcher with Technicolor, Rennes, France, from 2015 to 2017. My research interests include graph signal processing, graph-based machine learning and 3D visual computing. I have regularly published in top image processing / computer vision venues, including TPAMI, TIP, TSP, CVPR, ICCV, ECCV, etc., with more than 50 international journal and conference publications in total. I am the recipient of several paper awards, including Best Student Paper Runner Up Award in ICME 2020 and Best Paper Candidate in CVPR 2021. I was awarded the 2021 IEEE Multimedia Rising Star Award—Honorable Mention, ICME 21’ Outstanding Service Award, and the Peking University Boya Young Fellow. I have been a member of IEEE MSA-TC, and an Associate Editor for Signal Processing Magazine, IEEE Transactions on Signal and Information Processing over Networks and Frontiers in Signal Processing. I served as the Open Source Chair of ICME 2021, and an Area Chair of ICME 2020 and ACM MM 2020. Also, I co-organized special sessions in ICME 2020 and ICIP 2021, and a workshop in ICCV 2021. More details can be found on my professional website (https://www.wict.pku.edu.cn/huwei/). I love traveling, food and music.

My research lab, the GLab (Graph is Laplacian based), is dedicated to the research on graph signal processing (GSP), graph-based machine learning and 3D visual computing. The current projects involve robust and interpretable graph neural networks based on GSP, graph spectral processing and analysis of geometric data (e.g., point clouds), etc. The lab team consists of three PhD students, one master student, and several undergraduate students. We have constant collaborations with researchers worldwide, such as Prof. Gene Cheung from York University, Prof. Antonio Ortega from University of Southern California, Prof. Chia-Wen Lin from National Tsing Hua University, Dr. Anthony Vetro from MERL, etc.

2. What have been your most significant research contributions up to now?
Our most significant research contributions have been in the area of graph signal processing and graph-based machine learning. Three areas with the most significant contributions are: (a) Optimal graph Fourier transforms (GFTs) for image & point cloud coding. We proved optimality of (generalized) GFTs in terms of signal decorrelation assuming certain statistical processes, which is crucial for compact signal representation and leads to our proposed image / point cloud compression algorithms that outperform relevant MEPG standards, with many citations and impact in the industry. (b) Graph spectral processing of point clouds. We developed learnable graph filtering algorithms for the denoising, inpainting and resampling for both static and dynamic point clouds, leading to state-of-the-art performance and paper awards. (c) Unsupervised graph representation learning for point cloud analysis. We studied unsupervised graph representation learning, by developing Graph Transformation Equivariant Representation learning, etc., which is also more interpretable. Our methods push closer towards the upper bound set by the fully supervised counterparts.

3. What problems in your research field deserve more attention (or what problems will you like to solve) in the next few years, and why?

The emerging interpretable model-based geometric deep learning is expected to attract lots of attention in the research communities. The key problem we try to solve is robust and interpretable graph neural networks, for which model-based geometric deep learning is crucial. In particular, we aim to integrate the robustness merit of model-based approaches and the learning power of data-driven approaches to retain the benefits of both paradigms. Also, this can be extended to general model-based deep learning.

4. What advice would you like to give to the young generation of researchers/engineers?
a) It would be good to selectively read papers from various fields to know different research directions, which helps find your research interest;
b) Be critical to previous works, learn to think of ideas independently, and take actions to validate your ideas;
c) Discuss with your supervisors and labmates, and participate in relevant conferences to learn and exchange ideas.