Introduction

The theory and algorithms for statistical learning and data-driven algorithms have been around since the early 19th century. But the first hints of modern machine learning can be traced back to the 1943 work of McCulloch and Pitts [63] who proposed the first model of an artificial neuron loosely based on the functioning of a biological neuron in vertebrates. Arthur Samuel is popularly credited to have coined the term “machine learning” in 1959, when he was at IBM performing research on teaching a computer to play checkers [93]. Machine learning has been very successful in applications such as computer vision, speech recognition and natural language processing. But the last few years have also witnessed the emergence of machine learning (in particular deep learning) algorithms to solve physics-driven problems, such as approximating solutions to partial differential equations and inverse problems.