Quantum Generative Modelling Using the Kicked Ising Spin Chain Dynamics

Abstract

Quantum Machine Learning is a paradigm that aims to harness natural quantum dynamics to perform various machine learning tasks. Variational quantum algorithms, which form a hybrid classical–quantum framework, are a promising area of research, especially in the current NISQ era, where fully fault-tolerant quantum computation is not yet available. In this work, we replace the typical randomly parametrised circuits with the kicked Ising spin system in the Quantum Circuit Born Machine (QCBM), which is used as the ansatz. This choice is motivated by the structured dynamics of the system, which can potentially offer better expressibility with a reduced parameter space. We have treated it as a toy model and have analysed its performance for supervised generative tasks, such as learning analytical distributions and the Bars and Stripes (BAS) dataset. The MMD loss function is used for training purposes. Finally, we have presented the results observed and directions for improvement in its performance and future work.