B.S Student School of Electronic Engineering and Computer Science Peking University No.5 Yiheyuan Road, Haidian District, Beijing 100871, P.R.China Email: qixun.wang@pku.edu.cn [Github]

Bio

I am an first-year Ph.D. student in the School of Intelligence Science and Technology, Peking University.

Currently, I'm focusing on the Out-of-Distribution (OOD) Generalization of various machine learning models and scenarios, including Domain Generalization (DG) in CV tasks, graph ODD generalization and the generalization of LLMs.

Educations

B.S in In Intelligence Science and Technology, School of EECS, Peking University (PKU)

[2019.9 ~ 2023.6]

Ph.D. Candidate in Machine Learning and Computer Vision, School of Intelligence Science and Technology, Peking University (PKU) Advisors: Xianghua Ying , Yisen Wang

[2023.9 ~ 2028.9 (expected)]

Publications

Improving Out-of-distribution Robustness by Adversarial Training with Structured Priors (NeurIPS 2022 Spotlight, top 5% ). Qixun Wang*, Yifei Wang*, Hong Zhu, Yisen Wang [pdf] [slide]

Research Experiences

• Solving Domain Generalization via Adversarial Training with Structured Priors. [pdf] [slide] with Yifei Wang, Hong Zhu, advised by Yisen Wang.

  We investigate the limitness of OOD performance improvement of the conventional sample-wise Adversarial Training (AT) and empirically validate that the low-rank structure that lies in the perturbation of the Universal Adversarial Training (UAT) is beneficial to OOD improvement. We further propose two low-rank structured AT algorithms to alleviate this issue. We empirically and theoretically prove the effectiveness of our methods.

• Welcome to the paper reading list & notes of OOD Generalization & some other topics maintained by me! [link]

Selected Awards

• Yanchuang Capital Scholarship (top 6%) (2022)
• Merit Student at PKU (top 6%) (2022)
• Academic Innovation Award at PKU (top 1%) (2022)
• Award for Academic Excellence at PKU (2021)