Simulating quantum dynamics is considered to be intractable on a classical computer and the probabilistic properties of nature make it even harder to predict the outcome of a simulation. This thesis investigates the efficacy of machine learning techniques to solvethe simulation problem. In this work we investigate various neural network architectures and will attempt to predict the state of a quantum system in the following time step based on the current state.
We first examine and visualize multiple activation functions and test the effect on predictions. We then use Convolutional Neural Networks to find the ground state of a Transverse Field Ising Model by minimizing the energy, where the system has been parameterized for the paramagnetic and ferromagnetic phases, with near critical value for the transverse field in the ferromagnetic case. We then use Neural Ordinary Differential Equations to predict the time evolution of the system after an abrupt quench of the transverse field and evaluate the difference between different network types.
We initially split the simulation problem into two parts: Finding the ground state and predicting the time evolution. After that we combined both problems to produce a solution that results in a prediction of the future state of a Transverse Field Ising Model after a parameter quench of the Hamiltonian.

Keywords: Simulating Quantum Dynamics, Transverse Field Ising Model, Neural Ordinary Differential Equations