Abstract
Machine learning (ML) is a promising approach for performing challenging quantum-information tasks such as device characterization, calibration and control. ML models can train directly on the data produced by a quantum device while remaining agnostic to the quantum nature of the learning task. However, these generic models lack physical interpretability and usually require large datasets in order to learn accurately. Here we incorporate features of quantum mechanics in the design of our ML approach to characterize the dynamics of a quantum device and learn device parameters. This physics-inspired approach outperforms physics-agnostic recurrent neural networks trained on numerically generated and experimental data obtained from continuous weak measurement of a driven superconducting transmon qubit. This demonstration shows how leveraging domain knowledge improves the accuracy and efficiency of this characterization task, thus laying the groundwork for more scalable characterization techniques.
Highlights
Machine learning (ML) has recently been applied to solve problems in numerous areas of physics, including quantum-information science [1,2]
Since the true quantum trajectories are unknown and the experimental data is affected by nonidealities and noise beyond what can be captured with the physical model of Eq (3), quantifying the degree of success of the training is more difficult in this case
We have demonstrated that leveraging quantum mechanics in the design of a machine-learning approach makes the characterization of quantum dynamics more efficient and accurate when using limited experimentally available data
Summary
Machine learning (ML) has recently been applied to solve problems in numerous areas of physics, including quantum-information science [1,2]. By training a ML model to predict measurement outcomes from the weak-measurement data, we can characterize the qubit parameters via the quantum-trajectory formalism [21] This physics-inspired approach overcomes two important limitations of generic machine-learning tools that are agnostic to the physics of the problem at hand, such as Ref. [22], which considered learning quantum trajectories from the continuous weak measurement of a transmon qubit It is more efficiently trainable as significantly less experimental data is needed to train the ML model. The physics-inspired approach allows for a direct device characterization, whereas the black-box approach lacks interpretability as the trained ML model cannot be associated with an explicit physical description of the device We address these two limitations by designing ML models that exploit our domain knowledge of quantum trajectories and dispersive measurements in circuit QED [20].
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
