摘要: Adversarial vulnerability is the Achilles' Heel of modern deep learning. Understanding the mechanisms of adversarial attacks for deep learning helps the researchers to improve the trustworthiness of artificial intelligence. In this talk, we establish a connection between adversarial robustness and data geometry under the manifold hypothesis, and identify some fundamental causes of adversarial vulnerabilities.

报告人简介: 宋琪凡博士，于2009年获得北京大学统计学理学学士学位，于2014年获得美国德州农工大学统计学博士学位，现任美国普渡大学统计学副教授。宋教授的研究领域包括且不限于统计计算，贝叶斯渐近理论，高纬数据分析，深度学习等。宋教授的研究成果发表于统计期刊JASA，Biometrika，JRSSB，Bernoulli以及机器学习会议AISTAT，UAI和NeurIPS。