Siyan Zhao

I’m a 4th-year PhD student in Computer Science at UCLA. I am fortunate to be advised by Professor Aditya Grover.
My primary research interest lies in endowing machines with human-like reasoning and efficiency. My recent research focuses on:
- Understanding and scaling (diffusion) LLM reasoning via RL — d1, IGPO, and probing in-context learning.
- Efficient preference alignment & personalization — GPO, PrefEval.
- LLM inference efficiency & modular design for RL — Prepacking, Decision Stacks.
Prior to UCLA, I obtained my bachelor’s degree from the Engineering Science (Machine Intelligence program) at the University of Toronto. Before my PhD, I worked on 3D perception and RL algorithms for autonomous driving agents.
recent updates
[09/2025] | d1 is accepted at NeurIPS 2025 as spotlight. |
---|---|
[05/2025] | Gave invited talks on “Scaling Reasoning in Diffusion Large Language Models via Reinforcement Learning” at ASAP seminar, HKU NLP Group, UCSD Hao AI Lab, and NVIDIA! |
[02/2025] | PrefEval is accepted to ICLR 2025 as oral presentation. |
[09/2024] | Probing the Decision Boundaries of In-context Learning in LLMs was accepted at NeurIPS 2024 and also won the best paper award at the Foundation Model Interventions Workshop, NeurIPS 2024. |
[08/2024] | Excited to receive the 2024 Amazon Fellowship. |
selected works (all)
- Preprint