IBM demos quantum leap in artificial intelligence

MIT (Massachusetts Institute of Technology) and Oxford universities, along with researchers from IBM’s Q division, have published a paper detailing an experiment to show how quantum computing could accelerate machine learning.

The paper, published in Nature, an international journal of science, describes how quantum computers would be able to out-perform traditional computers for feature mapping. This is a component of machine learning that disassembles data to get access to finer-grain detail.

The technique of feature mapping in machine learning looks at known data to identify features in the dataset such as the differences between an image of a cat and a dog. Speaking to Computer Weekly about the paper, Kristan Temme, a researcher at IBM, said: “A feature map takes a bare datum and lists all the features it has.”

In machine learning, it is considered challenging to train a machine with complex data, where there there are less data samples than the number of features needed to uniquely identify a given thing, such as a cat or dog image.

For certain types of data analysis, the required computational resources needed for feature mapping scales exponentially with the size of the problem, making it hard to solve on a traditional computer. But this complexity appears to be an ideal fit for quantum computing, according to Temme.

“There is a natural confluence between feature maps and quantum mechanics,” he said. “You can apply a quantum circuit to a feature map that is arguably hard to do using classical machine learning on traditional computers.”

He added that while some feature maps work well on traditional computers, others perform better on a quantum computer. “We want to be able to identify feature maps that can’t be classified traditionally,” he said.

During the experiment, Temme said the researchers chose a feature map that was hard to process, and selected known dataset that the machine learning algorithm would be able to identify perfectly. The machine learning algorithm was then tested on a quantum computer to check it gave the predicted results.

Temme said the researchers need to identify more feature maps that can benefit from quantum computing. To aid this discovery, the entire software and access to the experiment has been made available as open source on GitHub, he added.

“We are taking a community approach. Many people will have to propose feature maps and data,” he said.

The quantum computer to run the experiment is also provided for free via IBM Q as a cloud service.