Modeling Long Short-Term Memory in Quantum Optical Experiments

Abstract

During the previous decade, artificial neural networks have excelled in a wide range of scientific disciplines, commercial applications, and everyday professions, including medical diagnostics, self-driving automobiles, and board games, to mention a few. In contrast to classic feed-forward neural networks, long short-term memory (LSTM) designs use recurrent connections to process sequential data such as text and speech. We explain how machine learning can be used to describe quantum physics experiments. Quantum entanglement is a key component of quantum technologies such as quantum computation and quantum cryptography. The study of complex quantum states with more than two particles and a large number of entangled quantum levels is particularly interesting. Reconstructing an experimental setup that yields such a multi-particle high-dimensional quantum state is usually impossible. To come up with interesting experiments, one must randomly generate millions of setups on a computer and calculate the resulting states. In this study, we show that machine learning models beat random searches by a significant margin. We show that without having to compute the states directly, an LSTM neural network can successfully train to simulate quantum experiments by correctly predicting output state characteristics for given settings. This strategy not only speeds up the search, but it's also a prerequisite for building multi-particle high-dimensional quantum experiments using generative machine learning models.