Monday 31 March 2025
Scientists have been working on a way to classify astronomical light curves, which are records of how bright stars and other celestial objects appear over time, using artificial intelligence. They’ve made significant progress by training visual transformers, a type of neural network, on large datasets of pre-processed images.
Light curves can be thought of as a series of snapshots showing the brightness of an object at different times. Astronomers use them to study everything from exoplanets to supernovae. However, with the sheer volume of data being collected, it’s becoming increasingly difficult for humans to manually analyze and classify these light curves.
To tackle this problem, researchers have developed a novel approach that involves transforming light curves into images. This is done by converting each point in the light curve into a pixel on an image grid, creating a visual representation of the data.
The team used a pre-trained visual transformer model called SwinV2 and fine-tuned it on their dataset. They found that this model was able to classify light curves with high accuracy, even when dealing with imbalanced datasets where one class has significantly more examples than others.
One of the key advantages of this approach is its ability to handle multi-band observations. Many astronomical surveys collect data in multiple wavelengths, which can be challenging to process using traditional machine learning methods. The visual transformer model can easily handle this type of data by treating each band as a separate image channel.
The researchers also explored different hyperparameters and found that the optimal settings varied depending on the specific dataset being used. For example, they discovered that higher learning rates worked better for smaller datasets, while lower learning rates were more effective for larger datasets.
This study demonstrates the potential of visual transformers in astronomical research and could have significant implications for our understanding of the universe. By automating the process of classifying light curves, scientists can focus on more complex and challenging problems, such as detecting rare events or studying the properties of distant galaxies.
The results are promising, but there is still much work to be done. The team plans to continue refining their approach and exploring its applications in other areas of astronomy. As our understanding of the universe continues to grow, it’s likely that we’ll see even more innovative uses of machine learning in astronomy.
Cite this article: “Classifying Astronomical Light Curves with Visual Transformers”, The Science Archive, 2025.
Astronomy, Artificial Intelligence, Light Curves, Neural Networks, Visual Transformers, Star Classification, Machine Learning, Exoplanets, Supernovae, Astronomical Surveys
