Original title: Recovering topological information of light by topological learning
Benquan Wang, Trishita Das, Yuhan Peng, Tatjana Kleine +7 · September 2026
Light waves with built-in topological patterns, like skyrmion numbers, are sturdy data carriers. But reading these patterns normally takes huge computing power. When light travels through messy, disordered paths, it scrambles into random-looking speckle patterns. This makes the hidden information seem completely destroyed and impossible to recover. To fix this, researchers built an artificial intelligence method called TOPO squared. The system does not need prior training. Instead, it looks directly at the mathematical topology inside the data across many length scales, matching the shape of the data to the shape of the light. It needs only one single intensity picture to work. The researchers found that this method accurately reads scrambled light states and outperforms standard algorithms. By using the skyrmion number to carry data through a disordered channel, TOPO squared correctly reconstructed the original transmitted images.
You have two identical-looking jigsaw puzzles that got mixed into one big pile. You cannot tell which piece belongs to which puzzle because they all look similar. But each puzzle has a unique curvature at the edges that stays the same even when jumbled. Here, the pile of pieces is the scrambled light speckle, the edge curvature is the topological invariant, and the AI inspects these curvatures across many small clusters to assign each piece to its original puzzle.
Imagine sending a message through a blizzard of interference—the message gets scrambled into visual noise. This paper shows that a specially designed AI can still read the hidden structure and reconstruct the original information. It means future optical communications could work reliably even under severe distortion, which is critical for high-capacity networks in hard-to-reach places.
The paper demonstrates recovery for light with a specific topological invariant (skyrmion number) through a disordered channel, but does not address whether the method generalizes to other types of topology or to dynamic channels that change in time. The approach also relies on computational extraction of multiscale features, which may have its own cost at very high resolutions or speeds. The abstract does not discuss noise beyond the speckle from disorder.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Recovering topological information of light by topological learning — Benquan Wang, Trishita Das, Yuhan Peng et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.