- Title： Improving Adversarial Robustness by Contrastive Guided Diffusion Process
- Date： 9:00pm US East time, Saturday, 11/12/2022
- Date： 10:00am Beijing time, Sunday, 11/13/2022
- Zoom ID：933 1613 9423
- Zoom PWD：416262
Title: Improving Adversarial Robustness by Contrastive Guided Diffusion Process
Synthetic data generation has become an emerging tool to help improve the adversarial robustness in classification. Among various generative models, the diffusion model has been shown to produce high-quality synthetic data. However, diffusion-type methods are typically slow in data generation as compared with other generative models. Although different acceleration techniques have been proposed recently, it is also of great importance to study how to improve the sample efficiency of generated data for the downstream task. In this paper, we first analyze the optimality condition of synthetic distribution for achieving non-trivial robust accuracy. We show that enhancing the distinguishability among the generated data is critical for improving adversarial robustness. Thus, we propose the Contrastive-Guided Diffusion Process (Contrastive-DP), which adopts the contrastive loss to guide the diffusion model in data generation. We verify our theoretical results using simulations and demonstrate the good performance of Contrastive-DP on image datasets.
Liyan Xie is an assistant professor in School of Data Science, The Chinese University of Hong Kong, Shenzhen. She received her Ph.D. in Industrial Engineering (major in Statistics) from Georgia Institute of Technology in 2021. Her research interests lie in the intersection of statistics and optimization, with a primary focus on sequential change detection and hypothesis testing, diffusion models for synthetic data generation, as well as their applications in sensor networks, healthcare, and spatio-temporal data processing.
Pleaes review our future speaker with this link: https://docs.google.com/spreadsheets/d/1qCV5u25mli2imCTXBD_Q9gWCvX6i_qR6/edit#gid=1787854274
Previous Talk from Dr.Junbin Gao, University of Sydney