Why this matters: instead of photographing a beam and crunching thousands of points digitally, this network lets light itself do the recognising — a shortcut that could make future optical links faster, tougher and less power-hungry.
October 14, 2026 — Researchers at the University of the Witwatersrand (Wits) have built a “light-powered brain”: an optical neural network that recognises hidden information in complex patterns of light without first taking a detailed picture. The work, published in Nature Communications and announced on October 13, comes from the Wits Structured Light Laboratory.
Image: A twisted ribbon of light passing through translucent optical layers toward glowing detectors. Credit: AI-generated illustration / Red Eye Eagle News, original illustration.
Editor's Highlight: The system reads a beam's topology — a hidden property that survives distortion — rather than its appearance, and identified 81 different light structures after training on very few examples.
Table of Contents
- Key Facts
- Counting the Twists in a Beam of Light
- How the Optical Network Works
- What the Tests Showed
- Why Communications Engineers Care
- The Limits of the Demonstration
- AI Meets Physics at Wits
- Bottom Line
- FAQ
- Sources
Key Facts
| Detail | Information |
|---|---|
| Institution | University of the Witwatersrand (Wits), Johannesburg, South Africa |
| Team | Supervised by Isaac Nape and Andrew Forbes; led by Hadrian Bezuidenhout, with Cade Peters and Ram Nandan Kumar |
| Publication | Nature Communications, October 7, 2026; announced October 13, 2026 |
| System | Deep diffractive optical neural network, two channels of five layers each |
| Result | 81 light structures recognised across 15 skyrmion values, at about 93% accuracy |
| Training | Nine simple modes per polarisation channel; training on just 15 examples |
| Potential uses | Optical communications, sensing, imaging and quantum communications |
Counting the Twists in a Beam of Light
Light can be shaped into special patterns, much as letters form words. Each pattern can act as a symbol in an optical alphabet, with no limit in principle to how many exist.
The patterns in this study are called optical skyrmions. Each carries a whole number — its skyrmion number — that stays the same even when the pattern is bent, scrambled or distorted along the way.
Dr Isaac Nape explains it with a ribbon: twist it, bend and fold it, and the number of twists never changes. Physicists call such hidden features topology.
How the Optical Network Works
The hard part is reading the number back. Conventional methods capture a detailed image and analyse thousands of points — a slow, computing-heavy process.
The Wits system instead passes light through two channels, each with five diffractive layers that steer it toward separate spots on a detector, according to TechCentral.
Where the light lands reveals the skyrmion number. In the lab setup, those layers are patterns on a spatial light modulator, a programmable device that reshapes light.
Newsroom Focus: The paper describes this as the first deterministic detector for optical skyrmions — giving a direct answer instead of an estimate from a full image.
What the Tests Showed
The network was trained live on the optical bench. By describing each layer with simple patterns instead of pixel by pixel, the team cut training parameters by a factor of about a thousand, TechCentral reports.
After learning nine simple modes in each polarisation channel, it identified 81 structures spanning 15 values, at about 93% accuracy — behaviour the authors compare to generalisation in ordinary neural networks.
The detector also held up under interference: with incoherent white light mixed in, accuracy stayed at 100% until the noise peak reached about 95% of the signal's intensity.
| Test | Outcome |
|---|---|
| Recognition | 81 structures identified across 15 skyrmion values |
| Accuracy | About 93% overall, per TechCentral's reading of the paper |
| Noise test | 100% accuracy until noise reached roughly 95% of signal intensity |
| Message test | An image transmitted and recovered using a 15-letter topological alphabet |
| Light throughput | About 15% of incoming light reached the detector in the lab setup |
The team also transmitted and recovered an image encoded in a 15-letter topological alphabet — a step from recognising patterns to carrying a message.
Why Communications Engineers Care
Analysis Note: Bigger alphabets mean more information per pulse — but only if the receiver can read them as fast as the sender creates them. That bottleneck is what this work attacks.
The approach could suit free-space laser links between buildings or to satellites, where turbulence scrambles beams. The authors also point to sensing, imaging and quantum communications, and say a trained detector could be built into fixed optics drawing no power.
Practical Takeaway: This is a measurement tool — the missing reader for light-based writing developed over years.
The Limits of the Demonstration
This is a laboratory demonstration, not a working link. Only about 15% of the light got through, mainly due to losses at the modulator and mirrors, which the authors say better components would reduce.
Earlier work from Forbes's group showed that entangled light's topology survives turbulence even when its patterns break down — this result addresses the reading half of that problem.
AI Meets Physics at Wits
Professor Benjamin Rosman, director of the Wits MIND Institute, says the work shows what becomes possible when AI and fundamental physics come together, embedding neural network principles directly into the behaviour of light.
Forbes, who directs the South African Quantum Technology Initiative (SA QuTI) and WitsQ, says creating these structures is only part of the challenge — practical ways to read them are still taking shape.
Nape says the team next wants to explore how optical intelligence could help quantum technologies in South Africa. Lead author Hadrian Bezuidenhout is now an optics engineer at APSI.
Bottom Line
A neural network built from light itself can read the hidden topology of structured beams, from few examples and under heavy noise. It is an early, laboratory-scale result — but it supplies the reader topological optical communications have been missing.
FAQ
What is a “light-powered brain”?
It is the university's name for an optical neural network — a system whose light-bending layers perform a recognition task instead of a computer program analysing an image.
What are optical skyrmions?
They are topological forms of structured light, each carrying a whole number that does not change when the pattern is distorted — making them robust information carriers.
How accurate was the system?
About 93% overall, according to TechCentral's account of the paper — and 100% in a noise test until interference reached roughly 95% of the signal's intensity.
When will this reach real communications networks?
There is no launch date. It is a laboratory demonstration with known limits, presented as a step toward future classical and quantum networks.
Who led the research?
It was conceived and supervised by Dr Isaac Nape and Professor Andrew Forbes, and led by MSc student Hadrian Bezuidenhout, with Cade Peters and Ram Nandan Kumar.
Sources
- Wits University — research announcement, October 13, 2026: https://www.wits.ac.za/news/latest-news/research-news/2026/2026-10/wits-scientists-create-light-powered-brain-that-could-transform-future-communications.html
- TechCentral — independent report with technical detail, October 13, 2026: https://techcentral.co.za/wits-light-powered-brain-optical-skyrmions/287258/
- Tech Xplore — report on the study and its Nature Communications details (DOI 10.1038/s41467-026-77376-5): https://techxplore.com/news/2026-10-optical-intelligence-communications-complex-patterns.html?deviceType=mobile
- Image: original AI-generated illustration created for Red Eye Eagle News.
Join the conversation
Comments
Post a Comment