Adversarial Machine Learning: Beyond Manipulating Pixels and Words

Explore the frontiers of adversarial machine learning in this thought-provoking lecture by Jerry Zhu from the University of Wisconsin-Madison. Delve into vulnerabilities in sequential machine learning, including multi-armed bandits and reinforcement learning, and discover how attackers can manipulate environments to force learners into adopting specific target policies. Examine the optimization of such attacks through control problems and higher-level reinforcement learning. Challenge the assumption that small pixel p-norm manipulations result in imperceptible attacks on image classification, and learn about a human behavioral study that questions the effectiveness of common metrics in matching human perception. Gain insights from Zhu's expertise as a Sheldon & Marianne Lubar professor, NSF CAREER Award recipient, and winner of multiple best paper awards, including an ICML classic paper prize.