Karen Aardal (Delft Institute of Applied Mathematics, Delft University of Technology)
Integer optimization and machine learning, some recent developments
In recent years the interface between machine learning and integer optimization has quickly been developing. We will give some examples of recent developments in the use of machine learning for learning how to branch and cut, and the use of integer optimization in adversarial machine learning.
Rémi Munos (DeepMind Paris)
Distributional reinforcement learning, Rémi Munos (Deepmind & Inria)
I'll talk about recent work related to the distributional reinforcement learning approach where the full return distribution is learnt instead of its expectation only. We generalize Bellman equations to this setting and describe a deep-learning approach for approximating those distributions. I will report experiments on Atari games.
Shai Shalev-Shwartz (Mobileye, an Intel company and Hebrew University of Jerusalem)
Teaching self-driving cars to be responsible: A mathematical formalism of the duty of care
An important element in making self-driving cars a real product, rather than a science project, is to provide clear safety guarantees. We argue that statistical guarantees give a very weak safety and propose instead a white-box, interpretable, mathematical model for safety assurance, which we call Responsibility-Sensitive Safety (RSS).