Safe Machine Learning with Applications to Control Theory

From movie/music recommendation to speech recognition, to forecasting the energy demand of consumers, machine learning techniques are nowadays at the heart of many systems and engineering tools. Sometimes, however, these same systems exhibit unexpected behavior that is not easily explainable. Whereas this is not a concern in many fields, it can lead to the failure of safe-critical systems such as autonomous vehicles, risking human lives. In this talk we will see how safety can be incorporated into machine learning, and how this can be applied to modeling and controlling dynamical systems.