Avatar

Yifei Wang

yifei_wang AT pku.edu.cn

Ph.D.

Peking University

Biography

I’m a Ph.D. student at the School of Mathematical Sciences, Peking University. Currently I work at ZERO lab led by Zhouchen Lin and I am co-supervised by Jiansheng Yang and Yisen Wang. I got my bachelor’s degree also from the School of Mathematical Sciences, Peking University.

Interests

  • Self-supervised Learning
  • Adversarial Robustness
  • Probabilistic Modeling

Education

  • Ph.D. candidate in Applied Math, 2019-present

    Peking University

  • Master student in Machine Perception, 2017-2019

    Peking University

  • BSc in Applied Math, 2013-2017

    Peking University

Internships

 
 
 
 
 

Disentangled Representations

Huawei Noah’s Arch Lab

Sep 2019 – Jul 2020 Beijing
 
 
 
 
 

AD Selection

Baidu’s Phoenix Nest

Sep 2018 – Feb 2019 Beijing

Publications @ZERO Lab

Optimization-induced Implicit Graph Diffusion. ICML, 2022.

Due to the over-smoothing issue, most existing graph neural networks can only capture limited de- pendencies with their inherently …
Qi Chen , Yifei Wang , Yisen Wang , Jiansheng Yang , Zhouchen Lin
PDF

Reparameterized Sampling for Generative Adversarial Networks. ECML-PKDD, 2021.

Recently, sampling methods have been successfully applied to enhance the sample quality of Generative Adversarial Networks (GANs). …
Yifei Wang , Yisen Wang , Jiansheng Yang , Zhouchen Lin
PDF

Demystifying Adversarial Training via A Unified Probabilistic Framework. ICML workshop 2021, 2021.

Adversarial Training (AT) is known as an effective approach to enhance the robustness of deep neural networks. Recently researchers …
Yifei Wang , Yisen Wang , Jiansheng Yang , Zhouchen Lin
PDF

Dissecting the Diffusion Process in Linear Graph Convolutional Networks. NeurIPS, 2021.

Graph Convolutional Networks (GCNs) have attracted more and more attentions in recent years. A typical GCN layer consists of a linear …
Yifei Wang , Yisen Wang , Jiansheng Yang , Zhouchen Li
PDF

Residual Relaxation for Multi-view Representation Learning. NeurIPS, 2021.

Multi-view methods learn representations by aligning multiple views of the same image and their performance largely depends on the …
Yifei Wang , Zhengyang Geng , Feng Jiang , Chuming Li , Yisen Wang , Jiansheng Yang , Zhouchen Li
PDF

Awards

Best (Student) Machine Learning Paper Award

ECML-PKDD 2021 Sep 2021

Silver Best Paper Award

ICML 2021 AML workshop Jul 2021