DNNsolver: Accurate and Efficient Deep Learning Modeling of Polarization-Sensitive Diffractive Neural Networks

Shuyi Wang; Hong-Bo Sun; Linhan Lin

Home/Archive/arXiv

arXiv · 2609.27323

DNNsolver: Accurate and Efficient Deep Learning Modeling of Polarization-Sensitive Diffractive Neural Networks

Shuyi Wang·Hong-Bo Sun·Linhan Lin

View original source ↗Read source document ↗Export metadata ↓

Abstract

Diffractive neural networks (DNNs), composed of cascaded diffractive layers, offer a promising platform for high-speed optical computing. However, accurate and efficient modeling of high-density, polarization-sensitive DNNs remains challenging. Conventional model based on thin element approximation (TEA) fails to capture electromagnetic coupling within layers, while full-wave simulations such as finite-difference time-domain (FDTD) methods are computationally expensive. Here, we propose DNNsolver, a deep learning-based electromagnetic surrogate model that learns the scattering response of diffractive layers. By predicting the scattering matrix rather than source-dependent output fields, DNNsolver decouples the incident wavefront from the model and could be applied to arbitrary input fields. A local scattering kernel and sliding-window strategy further allow efficient modeling of DNNs with arbitrary spatial sizes. For single diffractive layer under random speckle illumination, DNNsolver achieves an average mean squared error (MSE) of 0.0095 and about 10^6-fold speedup compared with FDTD simulations. DNNsolver is further applied to the optimization of a trilayer polarization-multiplexed image classifier, which exhibits better agreement with FDTD simulations than its TEA-based counterpart. Our work provides an accurate and efficient framework for the design and optimization of high-density, polarization-sensitive DNNs.

Explore related subjects

physics.optics

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Shuyi Wang, Hong-Bo Sun, Linhan Lin. 2026-09-23. DNNsolver: Accurate and Efficient Deep Learning Modeling of Polarization-Sensitive Diffractive Neural Networks. https://arxiv.org/abs/2609.27323

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Paraxial diffusion-field retrieval. II. Fokker-Planck generalization of the transport-of-intensity equation

The transport-of-intensity equation (TIE), namely the continuity equation associated with a coherent paraxial optical wavefield, is widely used for phase retrieval. It is a second-order partial differential equation which may be solved for the phase of a coherent paraxial field such as a monochromatic scalar optical beam, given the intensity and longitudinal intensity derivative in a plane perpendicular to the optical axis. We show how the coherent flow associated with the TIE may be augmented by a diffusive flow associated with a scalar or tensor diffusion field. Such diffusive flow can arise via scattering from unresolved spatially random microstructure in an illuminated sample, blurring effects of an extended chaotic source that illuminates the sample, the resolution-reducing effect of shot noise in detected intensity images of the sample, and the sharpening effect (negative diffusion) associated with scattering from sharp sample edges. Augmenting the TIE's modeling of coherent flow with a diffuse-flow channel leads to a Fokker-Planck extension to this equation. Two different augmentations are obtained, using several complementary derivations. The inverse problems of phase retrieval and diffusion-field retrieval are then considered, for defocus-based imaging and mask-based imaging. When symmetric overfocus and underfocus images are used for phase retrieval, the diffusive term drops out and our Fokker-Planck formalism implies that any ensuing TIE-based phase-retrieval method needs no modification in light of our formalism. However, the same focal-series dataset---typically an infocus image, a weakly overfocused image, and a weakly underfocused image---may also be employed to access the additional channel of information associated with the Fokker-Planck diffusion field. Our formalism is applicable to visible light, x-ray, electron, and neutron imaging.

physics.optics↗

arXiv2026

Robust multichannel bulk transport in a time-reversal-invariant insulator-free photonic waveguide array

Ultracompact cladding-free waveguide arrays with zero inter-channel spacing and negligible crosstalk open a new avenue for high-density integrated photonic circuits. However, existing cladding-free waveguide arrays typically rely on conventional trivial bulk modes, making them highly susceptible to scattering losses at sharp bends or in the presence of obstacles and defects. To overcome this limitation, we theoretically propose and experimentally demonstrate a robust, crosstalk-free, and cladding-free photonic waveguide array based on chiral anomaly bulk states (CABSs) in photonic crystals. By interfacing distinct Dirac photonic crystals that host Dirac cones at different high-symmetry points (Γ and K) in the Brillouin zone and carefully engineering the boundary conditions, the boundary-induced CABSs in adjacent channels become effectively decoupled due to a large momentum separation, thereby eliminating inter-channel crosstalk. More importantly, we experimentally demonstrate that these crosstalk-free CABSs are robust to perturbations, including metallic obstacles, air defects, and sharp bends. We further extend the CABS-based waveguide array to two dimensions and demonstrate a cladding-free triangular resonator and a crosstalk-free waveguide crossing, both of which are previously unattainable. Our work establishes a new design paradigm for cladding-free, crosstalk-free, and ultracompact topological photonic devices, paving the way for robust, highly integrated photonic circuits.

physics.optics↗

arXiv2026

A conformally-Euclidean Line Element for evaluating color differences

Starting from our previously proposed line element and considering more ``surface color'' datasets, we derive a simplified version which matches experimental datasets equally well and resulted into a conformally-Euclidean line element, which is conceptually much simpler than any existing color difference metrics. The color difference is written as an Euclidean difference multiplied with a simple factor which depends on the luminance only. In a subspace with constant luminance, as considered by MacAdam, this factor becomes constant and the subspace is flat. The same holds for sufficiently large luminances. Based on this LE we derive perceptual coordinates $\left(A,l_{c},s_{c}\right)$ very similar to the CIELab $\left(L^{*},a^{*},b^{*}\right)$.

physics.optics↗