Abstract

I will briefly discuss where we can expect machine learning to impact long standing problems in fundamental theoretical physics (e.g. finding testable predictions from string theory) and why it seems reasonable to start addressing these questions now. To illustrate the potential impact machine learning techniques can have, I focus on our recent example (arXiv:1809.02612), where we automatise the construction of physical models, satisfying both experimental and theoretical constraints. I present a framework which allows the generation of effective field theories using Generative Adversarial Networks. We identify consistent examples generated by the machine which fall outside the class of data used for training. As a starting point, we apply this idea to the generation of supersymmetric field theories. In this case, the machine knows consistent examples of supersymmetric field theories with a single field and generates new examples of such theories. In the generated potentials we find distinct properties, e.g. the number of minima in the scalar potential, with values not found in the training data. I will comment on further applications of this framework in string theory and fundamental physics.