Speaker
Description
Recent researches suggest Geometric Quantum Machine Learning (GQML) using Equivariant Quantum Neural Networks (EQNN) as a potential solution to overcome the main challenges encountered in QML, such as trainability and generalization. EQNN injects inductive bias based on the prior knowledge of the underlying geometry in the dataset to build a more robust model against changes in the input data and increase generalization power. Despite the promising progress, the studies are still limited to theoretical tests, and the role of realistic hardware noise in EQNN training has never been explored in depth. This work studies the impact of hardware noise on EQNN and compares it with non-equivariant models. We train a simple EQNN on two toy datasets in the presence of different types of noises and analyse the relation between the final training performance and the action of noises on the quantum state.
