A non-algorithmic approach to "programming" quantum computers via machine learning

TL;DR

This paper introduces a machine learning-based non-algorithmic method to program quantum computers, enabling efficient, noise-robust quantum computations without traditional algorithm decomposition, demonstrated through entanglement estimation on IBM hardware.

Contribution

It presents a novel machine learning approach to directly 'program' quantum computers, bypassing traditional algorithm construction and enhancing robustness to noise and decoherence.

Findings

Successfully estimated quantum entanglement experimentally

Ported the method to IBM quantum hardware

Used hybrid reinforcement learning for training

Abstract

Major obstacles remain to the implementation of macroscopic quantum computing: hardware problems of noise, decoherence, and scaling; software problems of error correction; and, most important, algorithm construction. Finding truly quantum algorithms is quite difficult, and many of these genuine quantum algorithms, like Shor's prime factoring or phase estimation, require extremely long circuit depth for any practical application, which necessitates error correction. In contrast, we show that machine learning can be used as a systematic method to construct algorithms, that is, to non-algorithmically "program" quantum computers. Quantum machine learning enables us to perform computations without breaking down an algorithm into its gate "building blocks", eliminating that difficult step and potentially increasing efficiency by simplifying and reducing unnecessary complexity. In addition, our non-algorithmic machine learning approach is robust to both noise and to decoherence, which is ideal for running on inherently noisy NISQ devices which are limited in the number of qubits available for error correction. We demonstrate this using a fundamentally non-classical calculation: experimentally estimating the entanglement of an unknown quantum state. Results from this have been successfully ported to the IBM hardware and trained using a hybrid reinforcement learning method.