Current quantum computers make too many errors to ever be truly useful, but an artificial intelligence that can correct quantum mistakes could offer a solution.
All computers, whether ordinary or quantum, require error correction. The task is more complex in quantum computing because each qubit, or quantum bit, exists in a mixed state of 0 and 1, and any attempt to identify errors by directly measuring qubits destroys the data.
This means it is early days for quantum error correction. Google announced in July 2021 that its Sycamore quantum processor was able to detect and fix computational errors using a human-created algorithm, but the additional hardware needed to do that introduced more errors than it was able to fix. Now, at RWTH Aachen University in Germany and his colleagues have demonstrated that an AI can eradicate errors from a group of qubits.
Advertisement
Quantum computers will be able to solve some problems vastly faster than classical computers, but, in theory, a classical computer can still do everything a quantum computer can 鈥� if given huge amounts of resources and time. So, because quantum computers are rare and expensive, the researchers instead simulated the qubits of a small quantum computer on very powerful classical hardware.
Using this simulated computer, they created a quantum version of an autoencoder, a type of machine-learning algorithm that can be trained to spot and fix errors. 鈥淢achine learning in the classical world has brought so many benefits,鈥� says Cardarelli. 鈥淪o why not try to use quantum machine-learning techniques for this?鈥�
The researchers fed data into the quantum computer to see which errors it produced, creating pairs of correct and error-riddled data. They used these pairs to train the quantum autoencoder to spot frequent errors and identify how to reverse them.
Cardarelli says that the AI could be trained on a particular quantum computer and therefore learn to fix the unique types of errors it generates. 鈥淯nlike classical computers, quantum computers are very fragile,鈥� he says. 鈥淪o the type of error may even depend on temperature, humidity, and may change from day to day.鈥� Laboratories have even seen errors occur when a train passes nearby due to almost imperceptible vibrations, which the AI could learn to ignore, he says.
This tailored training offers powerful advantages, says聽 at quantum computing firm IonQ. 鈥淭his is almost like custom designing an error-correction code according to the characteristics of the particular hardware,鈥� she says. 鈥淚 think it could potentially lead to much more efficient and faster error correction, so the resource requirements could be reduced. All of these things need to be validated on real hardware before we get too excited, but I think it鈥檚 great that people are coming up with these ideas.鈥�
搁别蹿别谤别苍肠别:听
Join us for a mind-blowing festival of ideas and experiences. New Scientist Live is going hybrid, with a live in-person event in Manchester, UK, that you can also enjoy from the comfort of your own home, from 12 to 14 March 2022. .
