We study expressive power of shallow and deep neural networks with piece-wise linear activation functions. We establish new rigorous upper and lower bounds for the network complexity in the setting of approximations in Sobolev spaces. In particular, we prove that deep ReLU networks more efficiently approximate smooth functions than shallow networks. In the case of approximations of 1D Lipschitz functions we describe adaptive depth-6 network architectures more efficient than the standard shallow architecture.

We show that under certain conditions one-dimensional dielectric lattice possesses total omnidirectional reflection of incident light. The predictions are verified experimentally using Na3AlF6/ZnSe multilayer structure developed by means of standard optical technology. The structure was found to exhibit reflection coefficient more then 99% in the range of incident angles 0-86 • at the wavelength of 632.8 nm for s-polarization. The results are believed to stimulate new experiments on photonic crystals and controlled spontaneous emission.

Abstract-Photonic crystals based on silica colloidal crystals (artificial opals) exhibit pronounced stopbands for electromagnetic wave propagation and the corresponding modification of the photon density of states in the visible range. These spectrally selective features can be enhanced by impregnating opals with higher refractive materials like, e.g., polymers. Doping of these structures with dye molecules, semiconductor nanoparticles (quantum dots), and rare-earth ions provides a possibility to examine the challenging theoretical predictions of the inhibited spontaneous emission in photonic bandgap (PBG) materials. First experiments are discussed in which pronounced modification of spontaneous emission spectra and noticeable changes in decay kinetics were observed.

We describe generalizations of the universal approximation theorem for neural networks to maps invariant or equivariant with respect to linear representations of groups. Our goal is to establish network-like computational models that are both invariant/equivariant and provably complete in the sense of their ability to approximate any continuous invariant/equivariant map. Our contribution is three-fold. First, in the general case of compact groups we propose a construction of a complete invariant/equivariant network using an intermediate polynomial layer. We invoke classical theorems of Hilbert and Weyl to justify and simplify this construction; in particular, we describe an explicit complete ansatz for approximation of permutation-invariant maps. Second, we consider groups of translations and prove several versions of the universal approximation theorem for convolutional networks in the limit of continuous signals on euclidean spaces. Finally, we consider 2D signal transformations equivariant with respect to the group SE(2) of rigid euclidean motions. In this case we introduce the "charge-conserving convnet" -a convnet-like computational model based on the decomposition of the feature space into isotypic representations of SO(2). We prove this model to be a universal approximator for continuous SE(2)-equivariant signal transformations.