Lei Jiang

I am a PhD student in the Department of Computer Science at the University of Illinois Chicago since 2024, working with Prof.Natalie Parde. Prior to this, I earned my Bachelor’s degree from Sichuan University.

My research focuses on making large language model (LLM) training more efficient and better understood. I study how to make training more effective under limited computational budgets, particularly in supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). For example, I investigate how to identify the most informative training examples and reduce the computational cost of data selection.

I am also interested in building better process and outcome reward models, particularly process reward models that capture meaningful reasoning progress and provide effective learning signals without relying on expensive step-level annotations.

Beyond improving training methods, I study how LLMs encode and acquire information internally, using techniques such as linear probing to understand how training shapes their representations and behavior. Ultimately, I aim to connect insights into how LLMs learn with practical methods for training more efficient, reliable, and interpretable models.