Weijie Su

Weijie Su is an Associate Professor in the Department of Computer and Information Science at the University of Pennsylvania's School of Engineering and Applied Science. His research focuses on machine learning, artificial intelligence, differential privacy, statistical theory, optimization algorithms, and data science. He explores foundational questions in large language models (LLMs), alignment mechanisms, privacy-preserving techniques, and algorithmic optimization. His work bridges theoretical insights with practical applications in AI ethics, peer review systems, and decentralized learning.

Notable research themes include analyzing LLM statistical foundations, developing matrix-gradient optimizers (e.g., PolarGrad), and addressing challenges in copyright and privacy for generative AI. He has contributed to frameworks for calibration-aware fine-tuning, magnetic preference optimization, and watermarking strategies. His studies also extend to game-theoretic alignment limits, communication-efficient optimization over manifolds, and bias mitigation in peer review.

His publications span top conferences and journals, emphasizing interdisciplinary approaches to AI safety, differential privacy, and algorithmic robustness. He actively explores theoretical underpinnings of deep learning dynamics, including neural collapse under privacy constraints and the interplay between optimization and stability.