Feature selection is a key feature in quantum annealing-based computing that can benefit machine learning, D-Wave CEO Alan Baratz argues.

D-Wave’s quantum annealing and classical computing hybrid systems are capable of optimizing feature selection, which the vendor says is the fundamental building block of machine learning.

“Whenever you do machine learning, your data set has a bunch of characteristics and what you're trying to do is build a model that represents those characteristics so when you present a piece of data it can determine whether that data matches those characteristics or does not match those characteristics. That's really what machine learning is about,” Baratz said.

The challenge is the large number of characteristics is huge and the extensive time it takes to train the model on those characteristics, he added. Feature selection is the process of identifying a small subset of characteristics in a data set that is highly representative of what the machine learning model is trying to learn.

This is a quadratic optimization problem with two variables interacting with one another, “but there are no good classical solvers for solving quadratic optimization problems. They're all designed for linear optimization problems,” Baratz pointed out. “And our quantum computer is very good at solving it.”

Quantum Annealing Vs. Gate Model

D-Wave is a pioneer in quantum annealing-based computing — also known as quantum annealing — which is an optimization process using quantum fluctuations to find the best solution. Gate-model quantum computers offered by Google, IBM, and other quantum computing vendors require problems to be expressed in terms of quantum gates.

“There is a lot of hope that a gate-model system will allow you to model training [machine learning] better than classical systems, basically because you're training on a quantum distribution rather than a classical distribution, which has many more degrees of freedom,” Baratz said. “There's a lot of research going on and a lot of experimentation going on, but nobody's yet proven it.”

D-Wave recently made its Quantum Feature Selection for Machine Learning service available on the Amazon Web Service (AWS) Marketplace. That service is designed to perform feature selection by constructing and solving a combinatorial optimization problem. The vendor claims that by using the quantum annealing-based computers a user can avoid the potentially costly cycle of iterative model training and incremental construction of the solution compared to the traditional feature selection methods such as recursive feature elimination.

The vendor is currently working with customers in finance on fraud detection, using the feature selection to find the right subset of data to train on for machine learning applications.