Jiancong Xiao

Welcome to my homepage! I am currently a postdoctoral researcher at the University of Pennsylvania, working with Prof. Qi Long and Prof. Weijie J. Su.
Privously, I obtained my Ph.D. from The Chinese University of Hong Kong, Shenzhen, where I was advised by Prof. Zhi-Quan (Tom) Luo and worked closely with Prof. Ruoyu Sun. Prior to that, I received my M.S. degree from The Chinese University of Hong Kong and my Bacholar’s degree from Sun Yat-sen University.
Research Interest: I am broadly interested in statistical learning theory and deep learning theory, with a focus on developing responsible and trustworthy machine learning models. My recent research focuses on statistical topics in large language models (LLMs).
-
Bias in LLM Alignment: Analyzing algorithmic bias and fairness in preference alignment methods. Developing preference alignment algorithms.
-
LLM Training and Fine-Tuning: Establishing theoretical foundations for fine-tuning and post-training algorithms. Addressing calibration issues in LLMs.
-
Adversarial Robustness: Developing theoretical frameworks for understanding adversarial examples, robust overfitting, and adversarially robust generalization.
-
Classical Learning theory: Studying Generalization (Rademacher complexity, Uniform stability, Pac-Bayes), Optimization (non-convex, non-smooth problem).
News
Mar 17, 2025 | Our paper titled “Statistical Impossibility and Possibility of Aligning LLMs with Human Preferences: From Condorcet Paradox to Nash Equilibrium” is available on arXiv now. |
---|---|
Jan 22, 2025 | Three papers on fine-tuning and preference optimization have been accepted to ICLR 2025. Congratulations to all the collaborators! |
Oct 07, 2024 | Attending COLM2024 at UPenn. |
Sep 20, 2024 | I will attend SIAM Conference on Mathematics of Data Science (MDS24) in Atlanta, Georgia. |
Aug 03, 2024 | I will attend JSM 2024 in Portland, Oregon. Welcome to my talk about algorithmic bias of LLMs in the session titled Harnessing Large Language Models: Opportunities and Challenges for Statistics. |
Jun 30, 2024 | I will attend COLT 2024 in Edmonton, Canada. I will chair a session titled “Adversarial/Robust Learning” and also present our recent work on adversarially robust generalization. |
May 08, 2024 | Two papers about adversarial training theory are accepted to COLT 2024 and ICML 2024, respectively! |