ﻻ يوجد ملخص باللغة العربية
Recently quantum neural networks or quantum-classical neural networks (QCNN) have been actively studied, as a possible alternative to the conventional classical neural network (CNN), but their practical and theoretically-guaranteed performance is still to be investigated. On the other hand, CNNs and especially the deep CNNs, have acquired several solid theoretical basis; one of those significant basis is the neural tangent kernel (NTK) theory, which indeed can successfully explain the mechanism of various desirable properties of CNN, e.g., global convergence and good generalization properties. In this paper, we study a class of QCNN where NTK theory can be directly applied. The output of the proposed QCNN is a function of the projected quantum kernel, in the limit of large number of nodes of the CNN part; hence this scheme may have a potential quantum advantage. Also, because the parameters can be tuned only around the initial random variables chosen from unitary 2-design and Gaussian distributions, the proposed QCNN casts as a scheme that realizes the quantum kernel method with less computational complexity. Moreover, NTK is identical to the covariance matrix of a Gaussian process, which allows us to analytically study the learning process and as a consequence to have a condition of the dataset such that QCNN may perform better than the classical correspondence. These properties are all observed in a thorough numerical experiment.
The study of deep neural networks (DNNs) in the infinite-width limit, via the so-called neural tangent kernel (NTK) approach, has provided new insights into the dynamics of learning, generalization, and the impact of initialization. One key DNN archi
The Neural Tangent Kernel (NTK) has discovered connections between deep neural networks and kernel methods with insights of optimization and generalization. Motivated by this, recent works report that NTK can achieve better performances compared to t
We present a novel neural network Maximum Mean Discrepancy (MMD) statistic by identifying a connection between neural tangent kernel (NTK) and MMD statistic. This connection enables us to develop a computationally efficient and memory-efficient appro
The prevailing thinking is that orthogonal weights are crucial to enforcing dynamical isometry and speeding up training. The increase in learning speed that results from orthogonal initialization in linear networks has been well-proven. However, whil
Deep residual network architectures have been shown to achieve superior accuracy over classical feed-forward networks, yet their success is still not fully understood. Focusing on massively over-parameterized, fully connected residual networks with R