Progress in scaling up quantum computing implementations has led to a new set of technical challenges - development of classical control techniques to initialize and control these systems. Control problems in one such implementation, quantum dots defined in semiconductor heterostructures, currently rely on gross scale heuristics from experiments. Fortunately, machine learning techniques for pattern recognition and image classification, using deep and convolutional neural networks (CNN), have shown surprising successes for a computer-aided understanding of complex systems. I will present our work [1,2] on using these techniques to characterize states and charge configurations of quantum dots. Using this characterization networks, we can recast the problem of tuning quantum dot devices to a required state as an optimization problem. We propose a closed-loop system of experimental control using a large, labelled simulated dataset, CNN-based learner, and numerical optimization techniques to automatically tune the quantum dot device. I will describe the application of these techniques to experiments as well as outline future problems for larger systems to be tackled with machine learning.
[1] Kalantre et. al., arXiv:1712.04914
[2] Zwolak et. al.., PLOS ONE 13, e0205844 (2018), arXiv:1809.10018
Notes: Lunch served at 12:00 pm, talk at 12:15 pm.
