Yao Xiao

CS Ph.D. candidate @ UIUC

Google Scholar
GitHub
X
Linkedin

About Me

I am a final-year Ph.D. candidate in Computer Science at University of Illinois Urbana-Champaign, advised by Prof. Derek Hoiem.

I aim to build intelligent agents that plan under partial information, remember what matters, and draw on past experience to avoid repeating failures. More specifically:

  • Closed-loop Planning: How can models make decisions under partial information? Each decision must build on past observations while shaping what comes next.

  • Memory Representation: How can an agent compress its past into a compact, updatable state that preserves what matters and turns experience into reusable lessons?

  • Experience Retrieval: How do we recall useful past experiences while filtering out distractions?

I'm seeking full-time opportunities starting Summer 2027. If you're interested in my research or would simply like to connect, feel free to email me!

News

  • [Oct 8, 2026] I was recognized as a Top Reviewer for NeurIPS 2026.

Selected Work

For the full publication list, please refer to my Google Scholar.

NeurIPS 2026

ReToken: Improving Long-Context VLMs with Visual Retrieval Token

Yao Xiao, Reuben Tan, Zhen Zhu, Yuqun Wu, Jianfeng Gao, Derek Hoiem

NeurIPS, 2026.

[TL;DR]
Attention tells you where a VLM looks, so shouldn't it tell you which frames matter? It doesn't: the retrieval signal lives in the value space, and a single learned token is enough to read it out.
[Paper] [Code]
ECCV 2026🏅Spotlight

How to Teach Large Multimodal Models New Skills?

Zhen Zhu, Yiming Gong, Yao Xiao, Yaoyao Liu, Derek Hoiem

ECCV, 2026. Spotlight.

[Paper] [Code]
TMLR 2025🏅J2C

TextRegion: Text-Aligned Region Tokens from Frozen Image-Text Models

Yao Xiao, Qiqian Fu, Heyi Tao, Yuqun Wu, Zhen Zhu, Derek Hoiem

TMLR, 2025. J2C certification, also presented at NeurIPS 2026.

[TL;DR]
A simple, training-free approach to get region tokens directly comparable to text embeddings, achieving state-of-the-art zero-shot region-level understanding.
[Paper] [Code]

Blog

A casual blog for sharing thoughts and tracking my journey. No tech talk, just real life.

→ Full list

Teaching

Services

Co-organizer: UIUC Computer Vision External Speaker Series, 2025 - 2026.

Reviewer: CVPR, ICCV, ECCV, ICLR, NeurIPS, WACV, IEEE Access.

Last Updated: 10/8/2026, 1:57:00 PM