AI Deciphers Quantum Light Signals with Record Accuracy

Scientists are turning to artificial intelligence to tackle the complexities of quantum materials. In a recent study, researchers focused on classifying light signals emitted by nanobubbles on monolayers of tungsten disulfide (WS₂). Characterising these quantum emissions is crucial for technologies like quantum cryptography, but distinguishing between subtle spectral bands can be notoriously difficult.

To solve this, the team transformed raw signal data into colour images using a technique called Continuous Wavelet Transform (CWT). These images were then analysed by deep learning models known as convolutional neural networks (CNNs), which are algorithms designed to recognise visual patterns. The strategy proved highly effective.

Among the tested architectures, the VGG16 model achieved the highest overall mean accuracy of 99.4%. It achieved perfect results for spectrally distant bands and maintained 96.5% accuracy even when distinguishing between adjacent, hard-to-separate signals. Another model, Xception, demonstrated remarkable computational efficiency, converging quickly with minimal training data. This successful fusion of classical machine learning and quantum physics offers a robust framework for future quantum sensing and information systems.