Deep Learning-Based Prediction and Optimal Sequential Measurement of a Quantum Dot

Spin qubits defined in quantum dots are promising for creating a scalable quantum computer. However, they are time-consuming to characterise, and as the size of these systems increases, this task will become intractable without the aid of automation. We present a machine learning algorithm that decides where to measure next and demonstrate it operating on a real quantum dot device in real-time. The algorithm utilises a probabilistic deep-generative model to make reconstructions of a full current map given partial measurement and information theory to select the most informative measurements to perform next.
We demonstrate, for two different measurement configurations, that the algorithm outperforms standard grid scan techniques, reducing the number of measurements required by up to 4 times and the measurement time by 3.7 times.