Photo

Yu Sun

Email: yusun [at] cs.stanford.edu

My research focuses on an algorithmic framework called test-time training. Its core idea is that each test instance defines its own learning problem, with its own target of generalization. This is usually realized by training a different model on-the-fly for each test instance using self-supervision.

I am a postdoc at Stanford University, hosted by Carlos Guestrin, Tatsu Hashimoto, and Sanmi Koyejo. I completed my PhD in EECS at UC Berkeley, advised by Alyosha Efros and Moritz Hardt. During my undergrad at Cornell University, I worked with Kilian Weinberger.

For a complete list of papers, please see my Google Scholar.

* indicates equal contribution.

Learning to (Learn at Test Time)
Yu Sun*, Xinhao Li*, Karan Dalal, Chloe Hsu, Sanmi Koyejo, Carlos Guestrin, Xiaolong Wang, Tatsunori Hashimoto†, Xinlei Chen†
[paper] [code]

Test-Time Training on Nearest Neighbors for Large Language Models
Moritz Hardt, Yu Sun
ICLR 2024
[paper] [code]

Test-Time Training on Video Streams
Renhao Wang*, Yu Sun*, Yossi Gandelsman, Xinlei Chen, Alexei A. Efros, Xiaolong Wang
[paper] [website]

Test-Time Training with Masked Autoencoders
Yossi Gandelsman*, Yu Sun*, Xinlei Chen, Alexei A. Efros
NeurIPS 2022
[paper] [website]

Test-Time Training with Self-Supervision for Generalization under Distribution Shifts
Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A. Efros, Moritz Hardt
ICML 2020
[paper] [website] [talk]

Older Papers

* indicates alphabetical order.

On Calibration of Modern Neural Networks
Chuan Guo*, Geoff Pleiss*, Yu Sun*, Kilian Q. Weinberger
ICML 2017
[paper] [code]

Deep Networks with Stochastic Depth
Gao Huang*, Yu Sun*, Zhuang Liu, Daniel Sedra, Kilian Q. Weinberger
ECCV 2016
[paper] [code] [talk]