State of the Art Tuning
In this video we provide on outlook on the state of tuning in contemporary tuning experiments, review the most frequently used classes of methods and how they go beyond the basics we covered in this course.
Main takeaways
- Contemporary used ML tuning methods broadly belong into the following three classes: deep neural networks, reinforcement learning and generative models
- Key theoretical and experimental challenge is to fully automate tuning of a chip with 15 or more working qubits.
Further reading
Adversarial Hamiltonian learning for quantum dots in a minimal Kitaev chain: https://arxiv.org/abs/2304.10852
Contemporary example of how generative models for tuning/Hamiltonian learning are used in practice.
Advances in Automation of Quantum Dot Device Control: https://arxiv.org/abs/2112.09362
Very nice review paper about control and tuning of quantum dots.
Modern applications of machine learning in physical sciences: https://arxiv.org/abs/2204.04198
If you are interested in a more broad perspective on machine learning in quantum physics (beyond tuning) check out this forthcoming book.