Near Term Quantum Machine Learning: Variational Quantum Algorithms Trainability and Expressivity

Résumé de l'exposé

Quantum Machine Learning (QML) algorithms are promising candidates for actual application of quantum computing. In particular, Variational Quantum Algorithms (VQAs) based QML algorithms seem particularly relevant for near term application. In order to design powerful algorithms, one needs to characterize the expressivity of quantum circuits, i.e., how relevant the set of functions produced are useful for learning problems. In addition, one must be careful about the capacity to train such circuits, as training variational circuits using gradient based methods can easily suffer from vanishing gradient phenomena.

The QML team led by Pr. Elham Kashefi is a joint group between the University of Edinburgh School of Informatics and Laboratoire d’Informatique Paris 6 (CNRS) working on near term application of quantum learning theory. In this presentation, we will present different figures of merits for Variational QML Algorithms and their link with trainability through the lens of our team works, with a particular focus on the study of quantum Fourier models and the use of subspace preserving algorithms.

Ajouter l'événement à l'agenda (ics)