Liangyu Wang
Liangyu Wang's research focuses on efficient systems and optimization for large language models, including distributed training and inference, GPU systems, zeroth-order optimization, privacy-preserving training, and matrix-based optimizers.
Biography
Liangyu Wang is a Ph.D. candidate in Computer Science at King Abdullah University of Science and Technology (KAUST), advised by Professor Di Wang. His work has appeared at venues including COLM, NeurIPS, and ACL, with projects such as ZO2, DistZO2, FlashDP, and Canzona. During his Ph.D., he has also worked on large-scale LLM pretraining, including an internship at Aramco and research with the Alibaba Qwen Team. Before joining KAUST, he received his master's degree from The Chinese University of Hong Kong, where he worked on multimodal machine learning.
Expertise and Interests
Liangyu Wang's research interests include optimizing distributed training and inference of LLMs, improving multi-threaded and multi-stream scheduling, and enhancing privacy-preserving methods for LLMs. Currently, I am working on:
- Efficient reinforcement learning (RL) for LLMs reasoning
- Distributed training and inference of LLMs
- Efficient algorithm and infrastructure design for LLMs
- Efficient privacy-preserving methods