Approach Uses Artificial Intelligence to Predict Quantum Advantage

Researchers from the Moscow Institute of Physics and Technology (MIPT), Valiev Institute of Physics and Technology, and ITMO University created a neural network that learned to predict the behavior of a quantum system by “looking” at its network structure. This convolutional neural network (CNN) is designed specifically to learn from graphs. It can autonomously predict which network solutions will demonstrate quantum advantage.

On a graph, a quantum walk — an advanced tool for constructing quantum algorithms — is fundamentally different from a random (classical) walk. On some graphs, quantum walks are faster than their classical analogs. When they are faster, quantum walks can enable quantum algorithmic applications and quantum-enhanced energy transfer.

However, little is known about the possible advantages of arbitrary graphs that do not have explicit symmetries. For these graphs, one would need to perform lengthy simulations of classical and quantum walk dynamics to check whether a speedup occurs.

The researchers’ CNN observes simulated examples and learns which complex features of a graph lead to a quantum advantage. The CNN is thus able to identify graphs that exhibit quantum advantage without performing quantum-walk or random-walk simulations.

AI on the lookout for quantum advantages. Courtesy of Alexey Melnikov.

Creating quantum computers is costly and time-consuming, and the resulting devices are not guaranteed to operate faster than a conventional computer. Researchers need tools for predicting whether a given quantum device will have a quantum advantage before the device is built. The CNN is able to distinguish between networks and determine if a given network will deliver quantum advantage, thus making it a useful tool for building a quantum computer.

The researchers evaluated their approach for line and random graphs and observed that the CNN’s classification was always better than random guesses, even for the most challenging cases.

“It was not obvious this approach would work, but it did. We have been quite successful in training the computer to make autonomous predictions of whether a complex network has a quantum advantage,” professor Leonid Fedichkin said.

“The distinctive feature of our study is the resulting special-purpose computer vision, capable of discerning this fine line in the network space,” researcher Alexey Melnikov said.

One of the processes that quantum walks describe well is the excitation of photosensitive proteins, such as rhodopsin or chlorophyll. A protein is a complex molecule whose structure resembles a network. Solving a problem that formally involves finding the quantum walk time from one network node to another could shed light on what happens to an electron at a particular position in a molecule, where it will move, and what kind of excitation it will cause.

Compared with architectures based on qubits and gates, quantum walks are expected to offer an easier way to implement the quantum calculation of natural phenomena. The reason for this, the researchers said, is that the walks themselves are a natural physical process.

The research was published in the New Journal of Physics (www.doi.org/10.1088/1367-2630/ab5c5e).