On Robustness and Generative Modeling

Explore the intersection of adversarial robustness and generative modeling in this comprehensive lecture presented by Dr. Iacopo Masi at the USC Information Sciences Institute. Delve into the reinterpretation of robust discriminative classifiers as Energy-based Models (EBMs) and discover new insights into adversarial training dynamics. Gain a deeper understanding of robust overfitting, distinct features of different adversarial training methods, and their generative capabilities. Learn about a novel process to enhance these capabilities without explicit generative model training. Examine the potential of using off-the-shelf robust classifiers to support generative modeling through inversion techniques. Benefit from Dr. Masi's expertise in machine learning, computer vision, and biometrics as he shares his research on adversarial robustness, proactive defense against image manipulation, inverse problems, and generative AI in both vision and NLP domains.