Accelerating quantum progress through machine learning
A basic challenge in quantum computing is to tune and characterise qubits on an ever-expanding scale. We have developed machine learning methods for quantum technologies, which are able to learn how to do this more efficiently than even experienced humans. This requires moving beyond methods which demand large amounts of readily available data, because in quantum technologies the data are often sparse and costly to acquire.
The machine learning is required not simply to classify the measurements which have been taken but to decide what parameters to set next. It can identify the Pauli spin blockade necessary for readout of singlet-triplet qubits. Without being reprogrammed, the machine is able to learn how to tune different architectures, and to characterise the variability of nominally identical devices.
To meet the commercial need for the techniques developed in our laboratory, we founded a company. The product is now in use around the world, accelerating the development of practical quantum computing. Alongside automated calibration of quantum experiments and devices, machine learning has the potential to accelerate quantum science though discovering and optimising quantum experiments, simulating quantum systems, and analysing quantum data.
As scientists we have the responsibility and the privilege of advocating the responsible use of the progress to which we contribute. This calls for insight from science and wisdom from other disciplines to learn how together we can seek to promote human flourishing in times which seem to be increasingly subject to uncertainty.