I am a first-year PhD student at Department of Electrical and Computer Engineering, Princeton University. Previously, I obtained B.Sc. in Computer Science with First Class Honours from the Chinese University of Hong Kong, with a minor in Mathematics. My current research interest lies in the theory and provable algorithms for optimization and machine learning, with a general zeal for solving practical problems with rigorous mathematics. My hobbies include piano, soccer (football), photography, among others.
Publications
-
Escaping Saddle Points in Heterogeneous Federated Learning via Distributed SGD with Communication Compression.
Sijin Chen, Zhize Li, and Yuejie Chi, International Conference on Artificial Intelligence and Statistics (AISTATS), 2024. -
Non-Convex Joint Community Detection and Group Synchronization via Generalized Power Method.
Sijin Chen, Xiwei Cheng, and Anthony Man-Cho So, International Conference on Artificial Intelligence and Statistics (AISTATS), 2024. -
CIA-SSD: Confident IoU-Aware Single-Stage Object Detector from Point Cloud.
Wu Zheng, Weiliang Tang, Sijin Chen, Li Jiang, and Chi-Wing Fu, AAAI Conference on Artificial Intelligence (AAAI), 2021.
Projects
-
Second-Order Convergence of Distributed SGD with Communication Compression
May 2022 – presentWe designed a distributed SGD algorithm with a novel error-feedback mechanism for communication compression, and proved a high-probability bound for the convergence to second-order stationary points of the proposed algorithm by showing the saddle-escaping property with the coupling sequence technique. The proof managed to remove the commonly used assumptions on local objective similarity, making our algorithm able to accommodate the federated learning settings.
-
Provably fast algorithms for joint community detection and group synchronization
Jun 2021 – Dec 2021We proposed a generalized power method (GPM) with spectral initialization to solve a joint problem of group synchronization and community detection. We established an estimation error bound for the spectral initialization using random matrix and random graph arguments, and proved the linear convergence guarantee for GPM, ensuring a significantly lower time complexity than the state-of-the-art semidefinite relaxation method.
-
3D object detection based on point clouds in autonomous driving scenes
Jun 2020 – Nov 2020
We designed 3D convolutional neural network models for autonomous driving scenes. I proposed data augmentation methods for performance improvement and validated their efficiency for model training on the benchmark dataset KITTI.
Honors and Awards
- Gordon Wu Fellowship
- Hong Kong Government Scholarship
Curriculum Vitae
View the PDF version here.