Artificial Intelligence News

Breakthroughs improve quantum AI

June 26, 2023

(Nanowerk News) Groundbreaking theoretical evidence suggests that a technique called overparametrization improves performance in quantum machine learning for applications that thwart classical computers.

“We believe our results will be useful in using machine learning to study the properties of quantum data, such as classifying different phases of matter in quantum materials research, which is very difficult to do on classical computers,” said Diego Garcia-Martin, a postdoctoral researcher. at the Los Alamos National Laboratory. He is co-author of a new paper by the Los Alamos team on engineering at Natural Computational Science (“Theory of overparameterization in quantum neural networks”).

Garcia-Martin is working on research at the Laboratory’s Quantum Computing Summer School in 2021 as a graduate student from the Autonomous University of Madrid.

Machine learning, or artificial intelligence, typically involves training neural networks to process information — data — and learn how to complete assigned tasks. In short, a neural network can be thought of as a box with knobs, or parameters, that take data as input and produce an output that depends on the configuration of the knobs.

“During the training phase, the algorithm updates these parameters as they learn, trying to find their optimal settings,” says Garcia-Martin. “Once optimal parameters are determined, the neural network should be able to extrapolate what it learns from the training instance to new and previously unseen data points.”

Classical and quantum AI share challenges when training parameters, as algorithms can reach sub-optimal configurations in their training and crash.

A leap in performance

Overparametriization, a well-known concept in classical machine learning that adds more parameters, can prevent that deadlock.

The implications of overparametrization in quantum machine learning models were poorly understood until now. In the new paper, the Los Alamos team defines a theoretical framework for predicting the critical number of parameters by which a quantum machine learning model becomes overparametrized. At a certain critical point, adding parameters drives a jump in network performance and the model becomes much easier to train.

“By establishing a theory that supports overparametrization in quantum neural networks, our research paves the way for optimizing the training process and achieving improved performance in practical quantum applications,” explains Martin Larocca, lead author of the manuscript and postdoctoral researcher at Los Alamos.

By leveraging aspects of quantum mechanics such as entanglement and superposition, quantum machine learning promises much higher speeds, or quantum advantages, than machine learning does on classical computers.

Avoiding the pitfalls in the machine learning landscape

To illustrate the Los Alamos team’s findings, Marco Cerezo, senior scientist on the paper and quantum theorist at the Lab, describes a thought experiment in which a hiker searching for the tallest mountain in a dark landscape represents a training process. Pedestrians can only step in a certain direction and judge their progress by measuring their altitude using a limited GPS system.

In this analogy, the number of parameters in the model corresponds to the direction available for pedestrians to move, said Cerezo. “One parameter allows back and forth movement, two parameters allow lateral movement and so on,” he says. The data landscape will likely have more than three dimensions, unlike our hypothetical pedestrian world.

With too few parameters, hikers are unable to explore thoroughly and may mistake a small hill for the tallest mountain or get stuck in a flat area where any step seems futile. However, as the number of parameters increases, the walker can move in more directions in higher dimensions. What may initially appear to be local hills may turn out to be a high valley between the peaks. With additional parameters, climbers escape the trap and find the true peak, or the solution to the problem.

Source link

Related Articles

Back to top button