Quantum Engineering of Qudits with Interpretable Machine Learning

Abstract

Higher-dimensional quantum systems (qudits) offer advantages in information encoding, error resilience, and compact gate implementations, and naturally arise in platforms such as superconducting and solid-state systems. However, realistic conditions such as non-Markovian noise, non-ideal pulses, and beyond rotating wave approximation (RWA) dynamics, pose significant challenges for controlling and characterizing qudits. In this work, we present a machine-learning-based graybox framework for the control and noise characterization of qudits with arbitrary dimension, extending recent methods developed for single-qubit systems. Additionally, we introduce a local analytic expansion that enables interpretable modelling of the noise dynamics, providing a structured and efficient way to simulate system behaviour and compare different noise models. This interpretability feature allows us to to understand the mechanisms underlying successful control strategies; and opens the way for developing methods for distinguishing noise sources with similar effects. We demonstrate high-fidelity implementations of both global unitary operations as well as two-level subspace gates. Our work establishes a foundation for scalable and interpretable quantum control techniques applicable to both NISQ devices and finite-dimensional quantum systems, enhancing the performance of next-generation quantum technologies.