Martin Saveski

Bio

Hi! I am a PhD student at UC San Diego, advised by Prof. Xiaolong Wang.

My research interest mainly lies in deep learning and language models. Recently, I work on Test-Time Training, which lets each test instance define its own learning problem to facilitate its own generalization. This learning problem can be self-supervised, making better generalization possible at test-time without access to ground-truth labels.

From my study of Test-Time Training, I have become particularly interested in developing

Models that reframe generalization as learning and are adaptable during deployment;

Self-supervised methods that emerge from and scale with data and compute;

Systems that support such methods with high parallelism and low latency.

I am very grateful to have worked with and learned from many great mentors along the way. I spent an unforgettable summer as a visiting student at Prof. Tatsunori Hashimoto's group at Stanford, working with Dr. Yu Sun and other mentors and friends. In my undergrad years, I worked with Prof. Jingjing Li at UESTC.

Publications

^‡ indicates equal contribution.

Learning to (Learn at Test Time): RNNs with Expressive Hidden States

Yu Sun^‡, Xinhao Li^‡, Karan Dalal^‡, Jiarui Xu, Arjun Vikram, Genghan Zhang, Yann Dubois, Xinlei Chen, Xiaolong Wang, Sanmi Koyejo, Tatsunori Hashimoto, Carlos Guestrin

arXiv 2024

Paper Code

Learning to (Learn at Test Time)

Yu Sun^‡, Xinhao Li^‡, Karan Dalal, Chloe Hsu, Sanmi Koyejo, Carlos Guestrin, Xiaolong Wang, Tatsunori Hashimoto, Xinlei Chen

arXiv 2023

Paper Code

Interpretable Open-Set Domain Adaptation via Angular Margin Separation

Xinhao Li, Jingjing Li, Zhekai Du, Lei Zhu, Wen Li

ECCV 2022

Paper Code

Imbalanced Source-Free Domain Adaptation

Xinhao Li, Jingjing Li, Lei Zhu, Guoqing Wang, Zi Huang

ACM MM 2021

Paper Code

Vitæ

Full Resume in PDF.

Acknowledgement

The template of this website is forked from Martin Saveski's repo. Many thanks!