Research uses machine learning to bridge quantum reality gap

This groundbreaking research, published in Physical Review X, has found a method to diminish the 'reality gap' - the discrepancy between theoretical predictions and actual performance of quantum devices.

Associate Professor Natalia Ares. The potential of quantum computing is vast, spanning areas such as climate analysis, financial prediction, pharmaceutical development, and enhancing artificial intelligence. To harness this potential, it's essential to scale and integrate quantum devices, or qubits. A significant obstacle is functional variability, where similar devices perform differently, believed to be due to minuscule flaws in the materials used.

These micro imperfections in quantum devices, unmeasurable directly, lead to differences between simulated and real outcomes.

The team tackled this by employing a 'physics-informed' machine learning strategy to indirectly deduce these imperfections, based on their impact on electron movement within the device.

Lead Researcher, Associate Professor Natalia Ares (Department of Engineering Science, University of Oxford), explained using a metaphor: “Imagine playing ‘crazy golf’ where the ball's speed and direction upon exiting a tunnel may be unpredictable. However, with repeated trials, a simulator, and machine learning, we could enhance our predictions, thus reducing the reality gap.”

 The researchers examined output currents across various voltage levels on a quantum dot device. They compared the actual current to a simulated current, assuming no internal disorder. The diverse voltage settings allowed the simulation to identify a pattern of internal disorder that could rationalise the observations at every voltage level. This method combined mathematical, statistical, and deep learning techniques. Associate Professor Ares further compared this to placing sensors in the golf tunnel to track the ball’s speed, improving shot predictions despite not seeing inside the tunnel.

The model not only identified appropriate internal disorder profiles to explain current measurements but also precisely predicted necessary voltage settings for certain operational states of the device. Importantly, this model offers a novel way to measure the variability between quantum devices. This can lead to more precise forecasts of device performance and guide the development of optimal materials for quantum construction. It suggests methods to counteract the impacts of material flaws in quantum devices. David Craig, a Ph.D. student at the Department of Materials, University of Oxford, and co-author, likened this to inferring black holes' existence from their effects on nearby matter. "We've used straightforward measurements as stand-ins for the internal variability of tiny quantum devices. While the actual device is more complex than our model, this research highlights the benefits of using machine learning informed by physics to bridge the reality gap."

The paper titled 'Bridging the reality gap in quantum devices with physics-aware machine learning’ is available in Physical Review X.