Researchers from Google DeepMind, in collaboration with the Google Quantum AI team, have developed an AI decoder designed to identify errors in quantum computing. In an article published in Nature, the scientists detailed how machine learning surpasses traditional methods in detecting errors in qubits. In the same issue, an expert from Delft University of Technology provided an in-depth analysis of the team’s findings.
One of the principal challenges in building fully functional quantum computers remains error correction. Qubits are notoriously fragile, leading to frequent computational failures. The Google team proposed a novel solution with their AI decoder, named AlphaQubit.
In recent years, Google has been advancing its quantum computer, Sycamore, which utilizes multiple physical qubits to create a single logical qubit. This approach enables the execution of programs while simultaneously addressing errors. The newly introduced decoder is a deep neural network trained to detect errors using data generated by Sycamore.
During experiments conducted on Sycamore with 49 qubits, as well as on a quantum system simulator, hundreds of millions of error instances were generated. Sycamore was then re-tested using AlphaQubit to identify and correct these errors. The results demonstrated a 6% improvement in error correction during highly accurate yet slower computations and a 30% enhancement when employing faster but less precise methods. In tests involving 241 qubits, AlphaQubit delivered performance exceeding expectations.
The researchers believe that the application of machine learning could significantly accelerate the progress of quantum computing, allowing efforts to focus on addressing other unresolved challenges.
