In this article, I would like to introduce a study that uses deep learning to extend the duration of the quantum cart-pole problem and make the energy closer to the eigenvalue. This is a quantum cart-pole problem in which the force F on the pole is used as a variable to determine the reward, and the Q function is maximized to make the cart-pole last longer without increasing the energy eigenvalue. The Q function that defines this problem is expressed as the product of the time expectation of the Hamiltonian and the time sum of the reward, which is calculated using the force F on the pole as the reward and the wave function that represents the subsequent time evolution. As a result, it was reported that the cartpole survived for a time comparable to that of the logarithmic Gauss in the input moment distribution and wave function. Furthermore, the higher the order of the applied potential, the longer it survives compared to the other methods. This is a slightly different approach from quantum machine learning, but it is one example of the use of machine learning in quantum mechanics. It also shows the potential for quantum simulation in other systems. It could be applied to a technique called quantum feedback, which could be useful for making qubits last longer. https://arxiv.org/pdf/1910.09200.pdf

Introduction of paper:Deep Reinforcement Learning Control of Quantum Cartpoles.

Hikaru Wakaura