Talk by Prof. Neider at SAIL Lecture Series

Abstract of the talk

Neural networks have become a crucial component in many modern technologies, including image and speech recognition, natural language processing, and autonomous driving. However, as these systems increasingly impact our lives, ensuring that they function correctly and make decisions that align with human values is essential. This necessity has prompted the development of methods for verifying that neural networks satisfy specific properties, such as safety, robustness, and fairness. In this talk, we give a gentle introduction to neural network verification, focusing on two key aspects. First, we discuss typical properties that one may want to verify. Second, we present an overview of the most common verification techniques, including deductive verification and abstract interpretation. Our goal is to provide a compact yet comprehensive introduction to this burgeoning field, enabling participants to understand the challenges and opportunities in verifying neural networks and providing them with tools for developing more reliable and trustworthy AI systems.

More information

See the home page of the lecture series.