|
John Shi I am a Postdoctoral Researcher in the Electrical and Computer Engineering (ECE) department at Carnegie Mellon University (CMU).I obtained my Ph.D. in ECE in 2022, advised by Professor Jose Moura. My thesis title was "A Dual Domain Approach to Graph Signal Processing," which focuses on new models for interpreting Graph Signal Processing and Geometric Deep Learning. Prior to that, I received my B.S. degrees in Computer Engineering and Applied Mathematics in 2017 from the University of Maryland, College Park. My research interests are graph signal processing (GSP) theory, its relationship to discrete signal processing (DSP), its applications, and graph convolutional neural networks (Geometric Deep Learning).
|
|
Research
|
|
1) GSP = DSP+ Boundary Conditions-The Graph Signal Processing Companion Model
John Shi, Jose Moura arXiv, 2024 arXiv Introduces the GSP companion model, which directly connects GSP and DSP. Shows that one can develop new GSP concepts using DSP intuition using the GSP companion model. |
|
2) Graph Convolutional Neural Networks in the Companion Model
John Shi, Shreyas Chaudhari, Jose Moura ICASSP, 2024 Shows that Graph Convolutional Networks are theoretically equivalent to traditional CNNs in the GSP companion model. |
|
3) Graph Signal Processing: Dualizing GSP Sampling in the Vertex and Spectral Domains
John Shi, Jose Moura IEEE Transactions in Signal Processing, 2022 arXiv / IEEE Illustrates the GSP sampling operation in both domains: the vertex and spectral domains. |
|
4) Graph Signal Processing and Deep Learning: The Role of Convolution, Pooling and Topology
Mark Cheung, John Shi, Oren Wright, Yao Jiang, Xujin Liu, Jose Moura IEEE Signal Processing Magazine, 2020 arXiv / IEEE A fundamental overview of Graph Signal Processing and its relationship with Geometric Deep Learning. |
|
5) Edge Entropy as an Indicator of the Effectiveness of GNNs over CNNs for Node Classification.
Yao Jiang, John Shi, Jose Moura Asilomar Conference on Signals, Systems, and Computers , 2020 arXiv Introduced a novel metric, edge entropy, to measure the effectiveness of graph CNNs on graph datasets. |
|
6) A Dual Approach to Graph CNNs
John Shi, Wendy Summer, Jose Moura Asilomar Conference on Signals, Systems, and Computers , 2020 By representing the data in the spectral domain, one can classify graph data that normal graph convolutional networks in the vertex domain cannot. |
|
7) On Graph Convolution for Graph CNNs
Jian Du, John Shi, Soummya Kar, Jose Moura IEEE Data Science Workshop, 2018 IEEE We explore a graph CNN called Topology Adaptive Graph Convolutional Networks (TAGCN), which utilizes the GSP definition of graph convolution. |
Miscellanea |
Awards |
June 2023: ICASSP Rising Star, Rhodes, Greece.
May 2023: A.G. Jordan Award. Awarded to a single graduating ECE Ph.D. student who has combined outstanding Ph.D. thesis work with exceptional service to the ECE or Carnegie Mellon communities. 2017: Ron Strauss Teaching Assistantship. One of only seven undergraduate students to teach Calculus at University of Maryland. |
Teaching Experience |
Tutorials:
Nov 2023: Graph Signal Processing: A Foundational Approach, IEEE PES/IAS PowerAfrica Conference. Co-Instructor: June 2023: Graph Signal Processing and Geometric Learning: A Foundational Approach, ICASSP Short Course. Fall 2022: Carnegie Mellon University Course: 18-898D: Special Topics in Signal Processing: Graph Signal Processing and Geometric Learning. Teaching Assistant: Carnegie Mellon University Courses: 18-202: Mathematical Foundations for Electrical Engineers, 18-290: Signals and Systems University of Maryland – College Park Courses: MATH140, 141: Calculus I and II, ENEE140: Introduction to Programming Concepts for Engineers, ENEE150: Intermediate Programming Concepts for Engineers, ENEE244: Digital Logic Design |
|
Last Update: Jan 10, 2025 |