Machine learning contributes to better quantum computing

Researchers from the RIKEN Center for Quantum Computing have used machine learning to correct errors in quantum computers – an important step in making these devices usable – using an independent correction process that, although predictable, can correctly determine how to make the necessary corrections.

The research is published in a journal Physical review letters.

In contrast to classical computers, which work on bits that can only take the basic values ​​0 and 1, quantum computers work on “qubits”, which can take any superposition of computational basis states. In combination with quantum entanglement, another quantum entanglement that connects different qubits beyond classical methods, this enables quantum computers to perform completely new tasks, offering potential benefits for certain computational tasks, such as large-scale search, optimization problems, and cryptography. .

The main challenge to the use of quantum computers comes from the extremely weak nature of quantum superpositions. Indeed, small disturbances caused, for example, by the presence of space everywhere lead to errors that quickly destroy quantum superpositions and, as a result, quantum computers lose their edge.

To overcome this obstacle, sophisticated quantum error correction techniques have been developed. While they can, in theory, effectively reduce the impact of errors, they often come with a large overhead in device complexity, which is itself error-prone and can increase exposure to errors. As a result, complete debugging is often difficult.

In this work, the researchers used machine learning to find error correction schemes that reduce tool overhead while maintaining good error correction performance. So far, they’ve focused on autonomous quantum error correction, where an intelligently designed, artificial environment replaces the need to perform constant error-checking measurements.

They also looked at “bosonic qubit encodings”, which, for example, exist and are used in some of the most promising and widespread quantum computing machines based on superconducting circuits.

Finding candidates that perform well in the large search area of ​​bosonic qubit encodings represents a difficult task of optimization, where researchers deal with reinforcement learning, an advanced machine learning method, where an agent explores a potentially invisible environment in order to learn and implement its policy of action.

With this, the team found that the incredibly simple, approximate qubit encoding not only significantly reduced device complexity compared to other proposed encodings, but also outperformed its competitors in terms of its error correction capability.

Yexiong Zeng, first author of the paper, says, “Our work not only demonstrates the potential of using machine learning for quantum error correction, but may bring us a step closer to successfully implementing quantum error correction in experiments.”

According to Franco Nori, “Machine learning can play an important role in addressing the challenges of large-scale computing and optimization challenges. Currently, we are participating in a number of projects that combine machine learning, artificial neural networks, quantum error correction, and quantum error tolerance.”