Predicting the minimum control time of quantum protocols with artificial neural networks

TL;DR

This paper uses machine learning, specifically an autoencoder combined with k-means clustering, to estimate the minimal control time in quantum state transfer protocols, capturing landscape topologies and improving quantum control efficiency.

Contribution

It introduces an unsupervised neural network approach to estimate the minimal control time in quantum protocols, revealing landscape topologies without prior analytical solutions.

Findings

The neural network accurately estimates the MCT for Landau-Zener Hamiltonians.

The approach captures the control landscape's topological changes.

Limitations were observed with the simple architecture used.

Abstract

Quantum control relies on the driving of quantum states without the loss of coherence, thus the leakage of quantum properties onto the environment over time is a fundamental challenge. One work-around is to implement fast protocols, hence the Minimal Control Time (MCT) is of upmost importance. Here, we employ a machine learning network in order to estimate the MCT in a state transfer protocol. An unsupervised learning approach is considered by using a combination of an autoencoder network with the k-means clustering tool. The Landau-Zener (LZ) Hamiltonian is analyzed given that it has an analytical MCT and a distinctive topology change in the control landscape when the total evolution time is either under or over the MCT. We obtain that the network is able to not only produce an estimation of the MCT but also gains an understanding of the landscape's topologies. Similar results are found for the generalized LZ Hamiltonian while limitations to our very simple architecture were encountered.