Towards Verified Deep Learning

Towards Verified Deep Learning

Explore the intersection of formal methods and deep learning in this 55-minute lecture by Sanjit Seshia from UC Berkeley. Delve into emerging challenges in deep learning, focusing on verification and robustness. Examine local robustness, semantic adversarial analysis, and differentiable rendering. Investigate the application of formal methods to cyber-physical systems with machine learning components. Learn about retraining techniques and the Scenic probabilistic programming language for scenario description. Gain insights into verified deep neural networks and future research directions in this field.