Gaoxiang Luo
(I go by Leon)
Ph.D. CandidateGroup of Learning, Optimization, Vision, Healthcare, and X (GLOVEX)
Department of Computer Science and Engineering
University of Minnesota Twin Cities
luo00042{at}umn[dot]edu
Keller Hall 6-210Hey there! I’m Gaoxiang Luo, a 3rd-year Ph.D. candidate at the University of Minnesota Twin Cities, advised by Prof. Ju Sun and Prof. Aryan Deshwal. I work on generative modeling and decision-making under uncertainty. My research has been supported by Zscaler, DSI-MnDRIVE award and a stealth startup.
Generative models under scarcity: little data to learn what is likely, and few evaluations to learn what is good.
- Learning what is likely from little data (pre-training as distribution matching). MM-FM (CVPR 2026) exploits the union-of-manifold structures in foundation-model latents to design source and coupling that enables small data generative models.
- Learning what is good from few evaluations (post-training and inference-time adaptation). RLSR (NeurIPS 2026) trains LLMs to optimize the risk-coverage tradeoff directly, so they answer when likely right and defer to an expert otherwise. COM-BOM (EMNLP 2025) uses Bayesian optimization to choose which candidates are worth an expensive evaluation.
Previously: I received my M.S. in Computer and Information Science from University of Pennsylvania. I was a research intern at Adobe Research, Endeavor AI, Zscaler, and Cisco Research with four granted U.S. patents filed during these internships.
News
| Sep, 2026 | I will be presenting Aligning Language Models with Selective Prediction at NeurIPS 2026 in Atlanta (Wed, Dec 9, 2026 • 10:00 AM – 1:00 PM EST at Hall C1). Happy to chat onsite! |
|---|---|
| Jun, 2026 | My first-named-inventor patent Query System using Multiple AI Agents for Text-to-SQL and Text-to-Python was granted (No. 12,645,672 B2). This was filed during my internship at Zscaler in 2024. |
| May, 2026 | I will be presenting Flow Matching for Multimodal Distributions at CVPR 2026 in Denver (June 6 Saturday 16:45 Exhibit Hall A #170). Happy to connect for casual chat onsite! |
| May, 2026 | Our paper Aligning Language Models with Selective Prediction got accepted to ICML 2026 Agents in the Wild: Safety, Security, and Beyond Workshop! |
| May, 2026 | I will join Adobe Research as an research scientist intern this summer in the bay area, always happy to connect for coffee chats, hackathons, AI meetups — or weekend adventures and dog play dates! |
