Applying Machine Learning to Quantum-Dot Experiments: Generation of Training Datasets and Auto-tuning

Arrays of gate-defined quantum dots provide a promising platform for the realisation of quantum computers. With experimental efforts moving towards such arrays, a new control challenge presents itself - determination of appropriate regions in the gate voltage space to allow efficient control and manipulation of the electrons. In the past, this challenge has been tackled with heuristic approaches. Machine learning tools have emerged as a practical toolkit for automated heuristics. I will describe our efforts to enable machine learning based auto-tuning of quantum dot arrays. A prerequisite is the availability of a training data-set that can qualitatively model the observed current and charge-sensor outputs. We estimate capacitance and tunnelling models of arrays under the Thomas-Fermi and WKB approximations. We then describe the learning problems on these datasets and outline an architecture for auto-tuning.