Changzhi Xie

Changzhi Xie (谢昌志) is a PhD student in computer science at the University of Southern California, working on machine learning theory with advisor Mahdi Soltanolkotabi. He represented the People's Republic of China at the 2015 International Mathematical Olympiad in Chiang Mai, Thailand, winning a silver medal with 25 points as part of a Chinese team that finished second overall. He attended Yali High School in Changsha, Hunan, and was admitted to Tsinghua University's Institute for Interdisciplinary Information Sciences (the 'Yao class'). His research spans training dynamics of neural networks, implicit regularization, and generalization in overparameterized models such as matrix sensing.

Notes7

• PhD in computer science at USC (2019-2025) under advisor Mahdi Soltanolkotabi, specializing in learning theory.2025
• Won a silver medal for China at the 56th IMO (2015, Chiang Mai) with 25/42 points; the Chinese team placed 2nd behind the US.Jul 16, 2015
• Thesis proposal 'On the Dynamics of Learning Linear Functions with Neural Networks' (April 2025); committee: Mahdi Soltanolkotabi, Haipeng Luo, Robin Jia, Vatsal Sharan, Adel Javanmard.Apr 14, 2025
• Co-author on 'Implicit Balancing and Regularization: Generalization and Convergence Guarantees for Overparameterized Asymmetric Matrix Sensing' (Soltanolkotabi, Stöger, Xie), COLT 2023 / IEEE Trans. Inf. Theory 2025.Mar 24, 2023
• Attended Yali High School in Changsha, Hunan, where he and classmate He Jiafan (贺嘉帆) were the two Hunan members of China's 2015 IMO team; both were admitted to Tsinghua's 'Yao class' (Institute for Interdisciplinary Information Sciences).Jul 18, 2015
• Work accepted at COLT 2026: 'When Both Layers Learn: Training Dynamics' (Berk Tinaz, Changzhi Xie, Mahdi Soltanolkotabi).2026
• Admitted to Tsinghua University's Institute for Interdisciplinary Information Sciences (Yao class) after being scouted as a math talent; entered September 2015.Apr 23, 2015