Chris Hou

Zhichao (Chris) Hou / 侯智超

I am currently Ph.D student of Computer Science at North Caroline State University. I am fortunately advised by Prof. Xiaorui Liu . Before joining NCSU, I got my master degree in Applied Mathematics at Academy of Mathematics and Systems Science, Chinease Academy of Sciences, advised by Lingyun Wu. I received my B.E. in Applied Mathematics at Beijing Normal University, where I was advised by Huajie Chen and Li Cui.

I was a research intern at Institute for AI Industry Research, Tsinghua University, working with Wenbing Huang, exploring spatial-temporal & equivariant graph neural networks.

I am a big fan of basketball and I idolize Chris Paul and Allen Iverson. Our team won two champions in basketball match in BNU.

Email / CV / Github / Linkedin / Google Scholar

Research Interest

I'm interested in adversarial attacks and defenses and Graph Neural Networks.

Graph Neural Networks
Robustness on Graphs, Spatial-Temporal GNNs, Equivariant Geometric GNNs

Adversarial Robustness
Adversarial Robustness for Graphs, Vision and Languange

Research Experience

	Robust Graph Neural Networks via Unbiased Aggregation Ruiqi Feng, Zhichao Hou , Tyler Derr, Xiaorui Liu Department of Computer Science, North Carolina State University Preprint, 2023 Provided a unified view of L1-based robust graph signal smoothing to justify the improved and closely aligned robustness of representative robust GNNs Propose a robust and unbiased graph signal estimator to mitigate the estimation bias in L1- based graph signal smoothing Designed an efficient Quasi-Newton IRLS algorithm to solve the non-smooth and non-convex estimation problem with a theoretical convergence guarantee
	Can Directed Graph Neural Networks be Adversarially Robust? Zhichao Hou, Xitong Zhang, Wei Wang, Charu C. Aggarwal, Xiaorui Liu Department of Computer Science, North Carolina State University Preprint, 2023 Constructed a high-level directed message passing scheme for undirected GNNs, such as GCN, GAT and APPNP Designed a novel directed adaptive attack to test the robustness of our model Our directed MP scheme is significantly robust to graph adversarial attacks both in both transferable and adaptive sense
	Automated Polynomial Filter Learning for Graph Neural Networks Wendi Yu, Zhichao Hou, Xiaorui Liu Department of Computer Science, North Carolina State University Preprint, 2023 Investigated the potential of graph polynomial filter learning Proposed Auto-Polynomial to enhance the performance and generalization of GNNs Evaluated our proposed framework on both homophilic and heterophilic graphs
	Equivariant Spatio-Temporal Attentive Graph Networks for Physical Dynamics Simulation Zhichao Hou, Liming Wu, Jirui Yuan, Yu Rong, Wenbing Huang Institute for AI Industry Research, Tsinghua University AI Lab, Tencent NeurIPS, 2023 Designed a novel Equivariant Discrete Fourier Transform (EDFT) to extract periodic patt Leveraged equivariant geometric GNNs capture the spatial dependencies in molecular graph Constructed a novel equivariant attention-based encoder to model the temporal dynamics
	PathExpSurv: Pathway Expansion for Explainable Survival Analysis and Disease Gene Discovery Zhichao Hou, Jiacheng Leng, Jiating Yu, Zheng Xia, Lingyun Wu Operations Research Laboratory, Academy of Mathematics and Systems Science Computational Biology Laboratory, Oregon Health and Science University BMC Bioinformatics, 2022 PDF / Code Two-phase training: pretrained with prior info and trained with penalty out of prior Pathways Expansion: expanded the prior pathways based on trained weights Downstream analysis: performed single-gene survival analysis, enrichment analysis and recoverability testing for the expanded pathways to do the disease drivers discovery
	Single-Cell ATAC-seq analysis via Network Refinement with peaks location information Jiating Yu, Duanchen Sun, Zhichao Hou, Lingyun Wu Operations Research Laboratory, Academy of Mathematics and Systems Science Briefings in Bioinformatics, 2022 PDF / Code Constructed the relation matrix based on the cell-peak accessible relationships and peak-peak co-accessibility Aggregated information with the Network Refinement (NR) diffusion method Leveraged SCARP to improve cell clustering performance and reveal new significant cell subpopulations

Wasserstein Distributionally Robust Optimization (WDRO)
Zhichao Hou, Lingyun Wu
Operations Research Laboratory, Academy of Mathematics and Systems Science

Undergraduate Research, 2019

Studied intensively on WDRO in classical ML problems (classification, regression, MLE, MMSE)

WDRO achieved better generalization and robustness against in SVM, Gaussian MLE, KF

Approximation Algorithms in Optimal Transport Wasserstein Distance
Zhichao Hou, Li Cui
Computational Mathematics Laboratory, Beijing Normal University

Undergraduate Research, 2018

Studied Sinkhorn algorithm based on entropy regularization theory

Researched Gibbs-OT algorithm based on Boltzmann distribution

Implemented Sinkhorn and Gibbs-OT algorithm in mage registration problem

Industrial Experience