Introduction to Professor Gene Cheung and His Lab at York University
Q1. Could you briefly introduce yourself (and your University/Lab)?
A1. I received my Ph.D. degree in electrical engineering and computer science from the University of California, Berkeley in 2000. I spent some time in industry, working in Hewlett-Packard Laboratories Japan, Tokyo, from 2000 till 2009. I returned to academia as an assistant professor in National Institute of Informatics (NII) in Tokyo, Japan, in 2009, where I became an associate professor in 2014. I moved back to my hometown Toronto in 2018, where I am now an associate professor in York University, Toronto, Canada.
I have served as associate editor for several journals, including IEEE Transactions on Multimedia (2007–2011), IEEE Transactions on Circuits and Systems for Video Technology (2016–2017) and IEEE Transactions on Image Processing (2015–2019). I am a co-author of several paper awards, including the best student paper award in ICIP 2013, ICIP 2017 and IVMSP 2016, best paper runner-up award in ICME 2012, and IEEE Signal Processing Society (SPS) Japan best paper award 2016. I am a recipient of the Canadian NSERC Discovery Accelerator Supplement (DAS) 2019. I am a fellow of IEEE.
My research interests include 3D imaging and graph signal processing, and my lab is called Graph and Image Signal Processing (GISP) Lab. My lab currently has 5 graduate students and 2 post-docs, working on topics such as image and 3D point cloud restoration, graph learning, metric learning, graph sampling and signal reconstruction. We have a new co-edited book on “Graph Spectral Image Processing” by Wiley-ISTE. York University is the 3rd largest university in Canada, and is conveniently located in Toronto, the largest and most diverse city in Canada. Summers are wonderful here and winters aren’t that bad!
Q2. What have been your most significant research contributions up to now?
A2. Much of my work in the past 10 years has focused on graph signal processing and its application to practical problems, specifically for imaging. In particular, i) we proved optimality of several variants of graph Fourier transforms (GFT) in terms of signal decorrelation for certain Markov source models, leading to state-of-the-art image compression algorithms, ii) we designed variants of graph Laplacian regularizations for a range of inverse imaging problems, including image denoising, deblurring, point cloud denoising and super-resolution, and iii) we developed an eigen-decomposition-free graph sampling algorithm extending from Gershgorin Circle Theorem (GCT) with roughly linear time complexity, which can scale gracefully to large graphs. Recently, the last work has matured into a set of linear algebraic theorems that has important implications in other fields beyond signal processing, such as machine learning and convex optimization.
Q3. What problems in your research field deserve more attention (or what problems will you like to solve) in the next few years, and why?
A3. We strive to combine the elegance and interpretability of model-based approaches with the powerful learning abilities of deep neural networks (DNN) into model-based deep learning systems that are compact, explainable and robust to covariance shifts. Algorithm unrolling is one obvious example, but this is only the beginning. Graph models can play an important role here, given its generality while maintaining its intuitive and desirable spectral graph filter interpretation.
Q4. What advice would you like to give to the young generation of researchers/engineers?
A4. Research is a team sport; don’t go it alone! It also means that you must: i) bring a unique and valuable skill set to the table (black boxes do not count); ii) communicate effectively with your team members; iii) be a good teammate—share the burden during tough times, and celebrate together during good times. We are building our research family one member at a time.