Speaker: Yinpeng Dong Tsinghua University
Time: 2024-11-21 14:00-2024-11-21 15:00
Venue: FIT 1-222

Abstract:

In the past decade, machine learning and deep learning have witnessed significant progress and been widely applied for pattern recognition tasks, such as computer vision, natural language processing, etc. However, a significant drawback of these algorithms is the vulnerability to adversarial examples, which are maliciously generated by adding human-imperceptible perturbations to natural examples but can mislead a model to make erroneous predictions. To study this problem, adversarial machine learning emerges as a new gamut of technologies that study vulnerability of ML approaches and detect the malicious behaviors in adversarial environments. This talk will introduce our efforts in adversarial ML in both theory and practice. On one hand, I will introduce efficient adversarial attacks, which are developed to identify the vulnerability of deep learning models. One the other hand, I will introduce effective adversarial defense methods that improve model robustness under adversarial attacks. Furthermore, this talk will introduce the adversarial robustness benchmark ARES and real-world applications of adversarial ML.

Short Bio:

Yinpeng Dong is a Postdoctoral Researcher in the Department of Computer Science and Technology, Tsinghua University. He received his BE and PhD degrees from Tsinghua University in 2017 and 2022, advised by Prof. Jun Zhu. His research interest includes machine learning, deep learning, and especially the adversarial robustness of deep learning. Yinpeng has published over 50 papers in the prestigious conferences and journals, including TPAMI, IJCV, NeurIPS, ICML, CVPR. These papers have amassed more than 10000 citations. He has organized several adversarial machine learning workshops at ICML’21, AAAI’22, ICCV’23, etc. He received CCF Outstanding Doctoral Dissertation Award, Tsinghua Oustanding Postdoctoral Researcher, Microsoft Research Asia Fellowship, Baidu Fellowship, ByteDance Scholarship, etc.