CLASSICAL AND QUANTUM ALGORITHMS FOR ORTHOGONAL NEURAL NETWORKS

• Main Authors: MATHUR, Natansh, KERENIDIS, Iordanis, LANDMAN, Jonas
• Format: Patent
• Language: English; French
• Published: 01.12.2022

Orthogonal neural networks impose orthogonality on the weight matrices. They may achieve higher accuracy and avoid evanescent or explosive gradients for deep architectures. Several classical gradient descent methods have been proposed to preserve orthogonality while updating the weight matrices, but...