Scalable Verification and Validation

While machine learning holds great promise for quantum technologies, most current methods focus on predicting or controlling a specific quantum system. The subfield of meta-learning enables one to learn a variety of similar tasks and adapt to new tasks from little information. By utilising prior data from different systems, this approach can learn new systems with less data and can go beyond quantum device variability. This can significantly accelerate predicting, controlling, tuning or optimizing the system, a crucial requirement for prototyping quantum technologies.

In this paper, we meta-learn dynamics and characteristics of systems including closed and open two-level systems, as well as the Heisenberg model. We predict the dynamics of a new system from very few data points after having trained on different systems from the same unknown Hamiltonian class.  Based on experimental data of Loss-DiVincenzo spin-qubits hosted in a Ge/Si core/shell nanowire with different charge occupation, we predict characteristics i.e. g-factor and Rabi frequency for a new qubit using meta-learning.

The algorithm we introduce adapts state-of-the-art meta-learning methods for physics-based systems while introducing novel techniques such as adaptive learning rates and a global optimizer. We benchmark our method against, vanilla transformer, and multilayer perceptron and demonstrate lower prediction errors.