Neurosymbolic Synthesis for Trustworthy Machine Learning

In this 47-minute lecture from the Simons Institute's "Theoretical Aspects of Trustworthy AI" series, Osbert Bastani from the University of Pennsylvania explores neurosymbolic synthesis approaches for developing more trustworthy machine learning systems. Discover how combining neural networks with symbolic reasoning can address fundamental challenges in AI trustworthiness, including interpretability, verification, and robustness. Learn about theoretical frameworks and practical implementations that bridge the gap between deep learning's flexibility and the formal guarantees of symbolic systems.