According to foreign media reports, researchers from U.S. Army combat capabilities development command’s Army Research Laboratory and Tulane University have combined machine learning with Quantum Information Science (QIS) to reconstruct the quantum state of an unknown system using photon measurement.
QIS is a rapidly developing field, which makes use of the unique characteristics of micro quantum systems, such as single particle or atom of light, to create powerful applications in communication, computing and sensing. However, traditional technologies are either unable to achieve or inefficient.
“We want to apply machine learning to QIS because machine learning systems can make predictions based on sample data sets without explicitly programming a given task,” said Dr. Brian Kirby, a scientist at the army enterprise research laboratory. In recent years, machine learning has performed well in computer vision and other fields. Training machine learning algorithms on a large number of pre classified images, and then we can correctly classify new images that we have never seen before For example, banks often use machine learning systems to read the handwriting on cheques, although the system program has never seen these special handwriting before. This image classification is similar to reconstructing quantum states from measurements, the researchers said.
“In image recognition, machine learning algorithms try to determine whether an object is a car or a bicycle,” said Dr. sanjaya lohani, a researcher at Durham University. The machine learning system is also effective in finding the specific features of the implied data sources in the measurement data. In both cases, the input data can be seen as a two-dimensional array from which the machine learning system attempts to pick out specific features. “
In order to describe the unknown quantum system, the research team used quantum state tomography (QST) technology. The researchers prepared and measured the same unknown quantum system, and used the complex calculation process to determine the most consistent quantum system with the measurement results. However, researchers still need to develop methods to deal with classical information related to quantum information protocols. Professor Ryan Glasser of Durham University said, “the classical information processing needed to operate quantum information systems is often ignored in this field. With the maturity of research and capability, real-world deployment is just around the corner. These are the problems we need to solve. “
Recently, researchers have developed a system that can reconstruct quantum states and standards, in some cases better than methods that require more computing resources. “We realized that we could match the performance of existing systems in the original simulation, and we wanted to see if we could train the system to predict common errors and build resilience in response to common errors,” said onur danaci, a researcher at Durham University
To this end, the team simulated common sources of error in measurement, such as misaligned optical elements, and used them to train machine learning systems. When the measurement results were not only noisy, but also completely lost, the researchers further tested the system. Danaci said that the team’s method is superior to the traditional state reconstruction method in each case, and requires less computing resources.
Because the team can load all the expensive calculations into the training process ahead of time, the actual reconstruction needs less resources. Kirby said the researchers hope to deploy these pre trained portable quantum systems in small field devices in the future, such as UAVs or vehicles with limited hardware space.