In recent years, machine learning has seen rapid advances with increasingly large scale and complex data modalities, including processing images, natural language and more. As a result, applications of machine learning have pervaded our lives to make them easier and more convenient. Buoyed by this success, we are approaching an era where machine learning will be used to autonomously make increasingly risky decisions that impact the physical world and risk life, limb, and property. For example, machine learning may autonomously decide when cars should brake or swerve, how power should be allocated in smart grids, what treatments to recommend in some medical settings and much more.

In this talk, I will discuss how we can begin to understand and mitigate this risk. In particular, I will focus on how we can combine lessons learned from the unprecedented practical success of deep learning with approaches from statistical and probabilistic machine learning to make risk aware decisions in practice and at scale. I will show how one popular probabilistic method, Gaussian process regression, can be made to scale without approximation to millions of training examples for complex tasks despite traditionally being limited to thousands. Finally. I will discuss a number of examples where these tools are deployed successfully in practice, and conclude with a discussion of the most important problems and limitations I believe we have yet to face in this area.