Chen Cai「蔡晨」

I am an engineer in Google Search working on LLM foundations for video ranking. My interest for work is to develop systems to have fine-grained understanding of long-form videos. I also spend some time on automating scientific paper writing with Google Research Applied Science team.

I obtained Ph.D. in the CSE from the University of California, San Diego, advised by Yusu Wang. My research interests are Geometric Deep Learning and Topological Data Analysis. See slides of a talk given at Microsoft Research AI4Science and final defense talk slides to learn more about my research.

Email   /   Google Scholar   /   Twitter   /   Github   /   LinkedIn     

• My research on geometric deep learning focuses on the theoretical understanding of graph neural networks (GNN). I worked on 1) provably improving the expressive power of GNN over 1-WL (Weisfeiler-Leman), 2) analyzing oversmoothing issue of GNN via Dirichlet energy, 3) the convergence of powerful GNN in graphon model beyond GCN, and 4) the connection between MPNN and Graph Transformer.
• I also applied graph neural networks to tasks such as graph coarsening, molecular dynamics, graph classification, and property prediction in geophysics and material science. See slides for my thesis proposal talk for more details.
• I worked on knowledge graph embedding, recommender systems, WebAnswers, and RNA structure prediction during internships at Baidu Research, Amazon, Google, and Atomic.ai.

• I complied a paper list for equivariant neural networks. Any suggestions are welcome!

Local-to-global Perspectives on Graph Neural Networks
Chen Cai,
Ph.D. thesis.

On the Connection Between MPNN and Graph Transformer
Chen Cai, Truong Son Hy,  Rose Yu,  Yusu Wang 
The International Conference on Machine Learning (ICML), 2023.  
paper / code
Theoretical connection between local Message Passing Neural Network (MPNN) with virtual node and global Graph Transformer in terms of functional approximation.

Composition Design of High-entropy Alloys with Deep Sets Learning
Jie Zhang*,  Chen Cai*,  George Kim,  Yusu Wang,  Wei Chen 
npj (Nature Partner Journals) Computational Materials, 2022.  
paper / code
Novel application of DeepSets for High-entropy alloys (HEA) property prediction.

Generative Coarse-Graining of Molecular Conformations
Wujie Wang,  Minkai Xu,  Chen Cai,  Benjamin Kurt Miller,  Tess Smidt,  Yusu Wang,  Jian Tang,  Rafael Gómez-Bombarelli 
The International Conference on Machine Learning (ICML), 2022.  
paper / talk / code
Generative model of molecular conformations with coarse-graining and E(3) equivariant neural networks.

Convergence of Invariant Graph Networks
Chen Cai,  Yusu Wang 
The International Conference on Machine Learning (ICML), 2022.  
paper / talk / code
The first convergence analysis of high order graph neural network under the graphon model.

Equivariant Subgraph Aggregation Networks 
Beatrice Bevilacqua*,  Fabrizio Frasca*,  Derek Lim*,  Balasubramaniam Srinivasan,  Chen Cai,  Gopinath Balamurugan,  Michael M. Bronstein,  Haggai Maron 
International Conference on Learning Representations (ICLR), 2022. (Spotlight, 5% acceptance rate)
paper / talk / blog / code
We present a provably expressive graph learning framework based on representing graphs as multisets of subgraphs and processing them with an equivariant architecture.

Equivariant geometric learning for digital rock physics: estimating formation factor and effective permeability tensors from Morse graph
Chen Cai,  Nikolaos Vlassis,  Lucas Magee,  Ran Ma,  Zeyu Xiong,  Bahador Bahmani,  Teng-Fong Wong,  Yusu Wang 
International Journal for Multiscale Computational Engineering, 2022.  
Morse graph representation + SE(3) equivariance for property prediction.

Graph Coarsening with Neural Networks
Chen Cai,  Dingkang Wang,  Yusu Wang 
International Conference on Learning Representations (ICLR), 2021. 
paper / talk / code
Learning based graph coarsening method based on graph neural network.

Network Representation Learning: Consolidation and Renewed Bearing
Saket Gurukar*,  Priyesh Vijayan*,  Aakash Srinivasan,  Goonmeet Bajaj,  Chen Cai;,  Moniba Keymanesh,  Saravana Kumar,  Pranav Maneriker,  Anasua Mitra,  Vedang Patel,  Balaraman Ravindran,  Srinivasan Parthasarathy 
Transactions on Machine Learning Research, 2022.  Presented at KDD Deep Learning Day,&nnbsp2021. 
paper / code
Large scale benchmark for various graph embedding methods.

Sanity Check for Persistence Diagrams
Chen Cai 
ICLR Workshop: Geometrical and Topological Representation Learning, 2021. 
paper / code
A simple sanity check for persistence diagram on various machine learning tasks.

A Note on Over-Smoothing for Graph Neural Networks
Chen Cai,  Yusu Wang 
ICML Workshop: Graph Representation Learning and Beyond, 2020. 
paper / talk / code
Theoretical study of the oversmoothing in GNN. Simplify earlier proof via Dirichlet energy and handle more nonlinearities.

Elder-rule-staircodes for augmented metric spaces
Chen Cai,  Woojin Kim,  Facundo Memoli,  Yusu Wang 
SoCG (Symposium on Computational Geometry), 2020. 
Full version is accepted to SIAM Journal on Applied Algebra and Geometry, 2021. 
SoCG paper / SIAM paper / arxiv / slides / code
Mathematics and algorithms of Elder-rule-staircodes for augmented metric spaces.

Group Representation Theory for Knowledge Graph Embedding
Chen Cai,  Yunfeng Cai,  Mingming Sun,  Zhiqiang Xu 
NeurIPS Workshop: Graph Representation Learning, 2019. 
A unification of various knowledge graph embedding methods via group representation theory.

A Simple yet Effective Baseline for Non-attribute graph Classification
Chen Cai,  Yusu Wang 
paper / short paper / code / pyG implementation
ICLR Workshop: Representation Learning on Graphs and Manifolds, 2019. 
A simple yet effective baseline for non-attributed graph classification that works surprisingly well compared to graph neural networks.

• Software Engineer Intern, Search Quality, Google, Summer 2022
• Applied Scientist Intern, Try before Your Buy Team, Amazon, Summer 2021
• Research Intern, Cognitive Computing Lab, Baidu Research, Summer 2019

• Reivewer for AAAI 2021, ICLR 2021 workshop, Journal of Applied and Computational Topology
• Program Committee for ICLR 2022 workshop on Geometrical and Topological Representation Learning

• Course instructor for Introduction to Computer Programming in C++ for Engineers and Scientists
• Teaching assistant for computational geometry (Fall 2015, Fall 2016), graph theory (Spring 2016), data structure and algorithm (Spring 2018), Artificial Intelligence I (Spring 2019)