Source-linked AI summary
Recovering topological information of light by topological learning
Benquan Wang, Trishita Das, Yuhan Peng, Tatjana Kleine, Shanshan Chang, Jinhui Chen, Nilo Mata-Cervera, Chunyu Li, Kelin Xia, Andrew Forbes, Yijie Shen
TL;DR
Strong disorder makes conventional recovery of optical topology intractable by producing unrecognizable speckle. The paper introduces TOPO2, which uses topology-aware multi-scale representations to recover skyrmion information from single-shot intensity, outperforming PCA and supporting classification across standard models.
Problem
Decoding topological information becomes impractical when strong disorder scrambles amplitude, phase, and polarization into speckle, while existing AI methods often omit physical invariants.
Method
TOPO2 uses topological representations of speckle, including simplicial complexes and persistent-homology lifetimes across scales, to recover skyrmion information from single-shot intensity.
Results
TOPO2 outperforms PCA across all seven metrics, including a 4.0× Fisher-type ratio improvement (12.90 vs. 3.21) and a 56.5× Calinski–Harabasz index improvement (6162 vs. 109).
Takeaways & Limitations
Topological information encoded in structured light can remain recoverable from single-shot speckle intensity without reconstructing the optical field or using Stokes polarimetry.
Abstract
from arXiv · showhide
The evolution of modern-day communication networks towards optical solutions with enhanced capacity and robustness is driving interest in topological light waves, exploiting their stability against perturbations through a topological invariant, e.g., the skyrmion number. However, detecting the underlying topology remains a computationally intense process even under ideal conditions, becoming intractable after passing through strongly disordered channels, where the degradation into unrecognisable speckle appears to destroy the topology. Here, we propose and demonstrate a topology-enhanced artificial intelligence (AI) approach to recover and classify such apparently lost topological information by computationally leveraging topological invariants in the data across many length scales. By aligning the topological classification of information with the topology of light, our topology-enhanced learning protocol, termed TOPO$^{2}$, achieves highly efficient recognition of the topological states of light, even from speckle, without the need for any prior learning. Our approach outperforms benchmark tests against standard computational algorithms and has the benefit of requiring just a single intensity pattern as the input, facilitating single-shot operation. To demonstrate this, we leverage the skyrmion number as a robust data carrier of images through a disordered channel, using TOPO$^{2}$ to accurately reconstruct the transmitted images. This work synergises topological photonics and topological AI for unravelling hidden topological signatures in light, opening a pathway towards robust communications even in extreme disordered environments.
INTRODUCTION
Topological light is promising for robust optical communications, but decoding its invariant information becomes impractical when disorder scrambles fields into speckle. TOPO2 addresses this by recovering topology from single-shot intensity patterns without full vector-field reconstruction.
- Topological optical structures can carry information because their invariant remains stable against continuous propagation perturbations.
- Stokes-polarimetry decoding is multi-shot and becomes impractical in dynamic or strongly disordered media.
- Existing AI decoders often rely on local intensity or low-order statistics rather than explicitly extracting optical-field invariants.
- TOPO2 retrieves topological information directly from single-shot intensity measurements by aligning learning with the skyrmion number.
- TOPO2 recovers topological information from scattered speckle without Stokes polarimetry or full vector-field reconstruction.
A. Hidden topological information in disordered speckle
The skyrmion number is a deformation-stable invariant encoded by polarization textures, but strong disorder destroys its directly observable global organization. TOPO2 instead uses deterministic topology-dependent intensity statistics to recover it from single-shot speckle.
- The skyrmion number counts how many times the normalized Stokes field wraps around the Poincaré sphere and remains conserved under continuous deformations.
- A mismatch in the azimuthal phase winding of two orthogonally polarized Laguerre–Gaussian modes creates spatially varying polarization assigned a skyrmion number.
- Strong disorder makes conventional Stokes-polarimetry recovery intractable by scrambling the field into fully developed polarization speckle.
- Statistical isotropy causes local Skyrme density to fluctuate in sign, so its spatial mean vanishes and the net skyrmion number approaches zero despite local twisting.
- Because transmitted intensity is a deterministic function of the incident field, topology-dependent statistical correlations remain available for representations sensitive to global geometric organization.
- TOPO2 extracts multi-scale topological features from single-shot speckle intensity using persistent homology and recovers the skyrmion number without Stokes polarimetry.
B. Aligning topological learning representation with topological light
TOPO2 represents speckle through persistent topological features rather than local intensity, aligning data topology with the skyrmion invariant. Multi-scale feature lifetimes provide descriptors for recovering topological states from single-shot measurements.
- B. Aligning topological learning representation with topological light: TOPO2 uses persistent homology to represent speckle through connectivity and loop structure rather than pixel-level intensity.The representation is designed to retain topology-dependent geometric organization after disorder scrambles the optical field.
- B. Aligning topological learning representation with topological light: Strong disorder converts the signal into visually indistinguishable speckle, while TOPO2 maps the speckle into topology-aware representations for single-shot recovery.The figure contrasts conventional recovery failure with the topology-enhanced representation.
- B. Aligning topological learning representation with topological light: Betti numbers quantify connected components and independent loops as integer-valued invariants that remain unchanged under continuous deformation.β0 counts connected components, while β1 counts independent loops.
- B. Aligning topological learning representation with topological light: A Vietoris–Rips filtration grows balls across scales, adding simplices when pairwise distances satisfy the construction criterion.Sweeping the radius generates a nested sequence of complexes in which topological features are born and die.
- B. Aligning topological learning representation with topological light: Feature lifetimes form TOPO2’s multi-scale descriptor, with long-lived structures associated with incident topology and short-lived features with disorder-induced noise.In the example, a loop appears at r = 0.52 and disappears at r = 0.72.
C. Single-shot topological information recovery platform without Stokes polarimetry
The experimental platform alternates between full Stokes polarimetry for ground-truth labels and single-shot speckle recording for TOPO2 recovery. It tests the approach across multiple disordered-media configurations.
- C. Single-shot topological information recovery platform without Stokes polarimetry: A shared optical beam path supports characterization with full Stokes polarimetry and recovery from a single camera exposure.Characterization establishes ground-truth skyrmion labels; recovery sends the same structured beams through disorder for TOPO2 classification.
- C. Single-shot topological information recovery platform without Stokes polarimetry: Skyrmion beams are generated by coherently superposing two orthogonally polarized Laguerre–Gaussian modes encoded on a binary-amplitude DMD.The two modes are separated, recombined, and assigned circular polarization components.
- C. Single-shot topological information recovery platform without Stokes polarimetry: A flip mirror redirects the beam from the polarimetry module through a scattering stage for recovery measurements.This switches the platform between ground-truth characterization and single-shot speckle acquisition.
- C. Single-shot topological information recovery platform without Stokes polarimetry: The experiment uses liquid-crystal, ground-glass, and programmable perturbation media to vary scattering mechanisms and scrambling strength.The programmable medium spans weak to strong phase distortion and partial to complete amplitude absorption.
D. Training-free interpretable topological learning reveals hidden topology
TOPO2 converts speckle persistence diagrams into unsupervised, fixed-length topology-aware features and compares them with variance-based PCA. These features separate skyrmion classes substantially better across the reported clustering metrics.
- D. Training-free interpretable topological learning reveals hidden topology: Each 256×256 grayscale speckle image becomes a point cloud whose Vietoris–Rips persistence captures topology across intensity-derived distances.Pixels with similar grayscale intensities are placed close together, while sharp local variations remain separated.
- D. Training-free interpretable topological learning reveals hidden topology: The resulting topology-aware feature vector is computed without labels or supervised training, providing an unsupervised representation for downstream learning.The representation is constructed before any classifier is applied.
- D. Training-free interpretable topological learning reveals hidden topology: Topology-aware features form compact, well-separated clusters ordered by skyrmion number, whereas PCA projections show severe overlap among six classes.PCA retains dominant variance, but the topology-aware space exposes classwise structure in the same speckle patterns.
- D. Training-free interpretable topological learning reveals hidden topology: 4.0×: the Fisher-type ratio improves from 3.21 to 12.90, while the Calinski–Harabasz index improves by 56.5× from 109 to 6162.Across all seven reported metrics, topology-aware representations outperform PCA.
E. Encoding and decoding information with topology
The communication protocol encodes image pixels as discrete skyrmion numbers and decodes the resulting topology-dependent speckle patterns from single-shot intensity. TOPO2 enables high-fidelity image recovery and maintains an architecture-independent advantage over raw-speckle decoding across tested disordered channels.
- Encoding: Image pixels are quantized into three-bit values and mapped to Nsk = 0–7, forming an eight-symbol topological alphabet for transmission.Each symbol determines the structured optical field launched through the disordered channel.
- Decoding: After propagation, visually random speckle retains topology-dependent statistical structure, so decoding identifies each transmitted topological symbol rather than conventionally reconstructing the image.Repeating symbol identification pixel by pixel enables full image recovery.
- Experimental demonstration: TOPO2 reconstructs encoded images from single-shot speckle intensity, with high fidelity demonstrated for three representative images transmitted through a liquid-crystal disordered medium.The decoder classifies topology-aware feature vectors using a lightweight multilayer perceptron.
- Benchmark comparison: TOPO2’s topology-aware representation outperforms raw-speckle ImageNet-pretrained ResNet-34 across tested architectures, including support vector machines, random forests, XGBoost, and multilayer perceptrons.The reported architecture independence attributes discriminative power to the representation rather than the downstream classifier.
- Disordered channels: Across liquid-crystal, ground-glass, and programmable perturbation media, TOPO2 recovers topological information from a single-shot intensity pattern and outperforms the pretrained ResNet-34 baseline.The tested conditions cover only part of practical scattering regimes, while indicating that the pipeline is not tied to one scattering mechanism.
DISCUSSION
TOPO2 recovers topological information from single-shot speckle intensity after strong scattering by representing the speckle’s statistical geometry rather than reconstructing the optical field. This topology-aware approach supports optical information transmission while reducing the parameter count and training cost associated with conventional convolutional networks.
- DISCUSSION: Strong scattering redistributes, rather than eliminates, topological information into the statistical geometry of speckle fields recoverable from single-shot intensity.Recovery requires neither full optical-field reconstruction nor Stokes polarimetry.
- DISCUSSION: TOPO2 writes information into topology and reads it through a topology-aware representation, establishing a paradigm for optical transmission through complex media.The framework operates on global geometric features instead of raw intensity distributions.
- DISCUSSION: Global geometric features reduce the parameter count and training cost relative to conventional convolutional networks, supporting computationally light and resource-compatible implementation.The passage links this computational lightness to practical efficiency and physical implementation.