Mengxi Wu 武梦溪

I am a Ph.D. student in the Department of Computer Science at the University of Southern California, advised by Prof. Xuezhe Ma. I received my Master of Science in Computer Science from New York University, where I was advised by Prof. Yi-Jen Chiang, Prof. Christopher Musco, and Prof. Yi Fang. Before that, I was an undergraduate student in EECS at the University of Michigan, Ann Arbor.

Research

My research focuses on the theoretical foundations for improving efficiency and reducing the computational cost of training large language models (LLMs). Specifically, I am interested in developing new optimizers or improving existing ones (e.g., AdamW, Sophia). I also work on the theory of hyperparameter optimization by analyzing the mathematical relationships between hyperparameters (e.g., weight decay), model architectures, and dataset characteristics across different scales. Previously, my work included transfer learning with theoretical guarantees, adversarial machine learning for 3D point cloud processing, and streaming algorithms for time-varying volume data.

Publications and Manuscripts (By Years / Selected)
Curvature Diversity-Driven Nuclear-Norm Wasserstein Domain Alignment for Point Cloud
Mengxi Wu, Hao Huang, Yi Fang, Mohammad Rostami

arXiv, 2024

[Preprint]

Graph Harmony: Denoising and Nuclear-Norm Wasserstein Adaptation for Enhanced Domain Transfer in Graph-Structured Data
Mengxi Wu, Mohammad Rostami

Transactions on Machine Learning Research, 2024

[Paper] [Code]

Streaming Approach to In Situ Selection of Key Time Steps for Time-Varying Volume Data
Mengxi Wu, Yi-Jen Chiang, Christopher Musco

Eurographics/IEEE Conference on Visualization, 2022

[Paper] [Code]

3D Point Cloud Completion with Geometric-Aware Adversarial Augmentation
Mengxi Wu, Hao Huang, Yi Fang

International Conference on Pattern Recognition, 2022

[Paper] [Code]


Website template from Jiayuan Mao. Last updated: Mar 2025.