Efficient Learning of Response Functions for Analog Quantum Computers

Programming analog quantum computers can be challenging due to errors in the underlying hardware. For these systems to serve as reliable computing platforms, we need a response function that maps programmed inputs to specific properties of the observed output distribution. In this study, we introduce an efficient machine learning method to learn such response functions from the measurement data obtained from these systems. We have tested our algorithm on D-Wave and QuEra quantum computers, demonstrating its ability to accurately predict their output distributions. Our approach is based on methods proven to be sample-optimal for learning graphical models. This algorithm can operate in the background during system use and adapt in real-time, learning the response as users execute tasks unrelated/independent tasks on the quantum machine.