In this video, we motivate why machine learning algorithms are beneficial in tuning and control task of quantum devices - and specifically, those based on semiconductor technology. We cover the concept of tuning as a high-level optimization problem and show how to conceptualize it as an input-output machine learning problem.
- Any tuning problem can be understood as a high-dimensional optimization
- There are distinct levels of abstraction in tuning:
- directly parsing the structure in measured data
- mapping the measured data on physics parameters
- Solid-state qubits are unique candidates for ML techniques: inherent properties of semiconductor qubits present time-dependent complex noise that requires generalisation and more sophisticated data processing
True or False:Semiconductor qubits are easier to fabricate than superconducting qubits and therefore they are easier to tune.
A nice introduction to supervised learning with neural networks: https://ml-lectures.org/docs/supervised_learning_w_NNs/ml_supervised_w_NNs.html
Online book for Neural Networks and Deep Learning: http://neuralnetworksanddeeplearning.com/
Experimental paper that inspired the classification exercise we performed in this course: https://journals.aps.org/prapplied/abstract/10.1103/PhysRevApplied.13.054005 (Arxiv: https://arxiv.org/abs/1911.10709)