Title: Predicting quantum dynamical cost landscapes with deep learning

Abstract: State-of-the-art quantum algorithms routinely tune dynamically parametrized cost functionals for combinatorics, machine learning, equation-solving, or energy minimization. However, large search complexity often demands many (noisy) quantum measurements, leading to the increasing use of classical probability models to estimate which areas in the cost functional landscape are of highest interest. Introducing deep learning based modelling of the landscape, we demonstrate an order of magnitude increases in accuracy and speed over state-of-the-art Bayesian methods. Moreover, once trained the deep neural network enables the extraction of information at a much faster rate than conventional numerical simulation. This allows for on-the-fly experimental optimizations and detailed classification of complexity and navigability throughout the phase diagram of the landscape.