The application of machine learning (ML) techniques to physics is a very active research field. In particular, in the last decade, deep learning has set a new standard in computer vision and natural language processing and is now regularly used in physics. Recent developments in this field seek to find the physical interpretation of ML models for applications in knowledge discovery. In this thesis, we focus on knowledge discovery in stochastic processes and the application of ML to optimization problems. We aim to apply interpretability techniques to stochastic data, for example, Brownian motion and fractional Brownian motion trajectories. Such techniques include deep generative models such as variational autoencoders combined with autoregressive models. As a second direction, we propose ML-inspired methods to benchmark different optimization methods such as simulated annealing, as well as to accelerate the search of ground states on Ising problems. In particular, we will use neural network based classifiers and restricted Boltzmann machines.

Tuesday October 14, 10:00 h. Elements room

Thesis Director: Prof. Dr. Maciej Lewenstein