Accelerating heliophysics workflow with interpretable deep learning

As we cruise through the age of information, we face a steep rise in data availability, especially high-resolution data in astrophysics. Extracting physics from such large datasets is computationally intensive and conditional on assumed physical models. Hence, such large data is ideally suited to be used with Deep Learning (DL) methods for applications ranging from forecasting to classification to inversion. In this talk, I will primarily present three examples of deep learning massively accelerating workflow in the domain of heliophysics and space weather. These examples operate on remote sensing data from the Sun, in-situ particle measurements near the Earth, and geomagnetic field measurements on the ground, tackling simulation-based inference and space weather forecasting. We demonstrate the application of interpretable DL to extract physical associations from such DL models, while also showcasing the development of physics-incorporated DL models. We discuss potential applications and caveats in using DL techniques for inferring physics hitherto unknown while also discussing how DL potentially eases multiple aspects of heliophysics workflows - especially through open-source workflows. I shall conclude with some of our efforts across the globe on the development of models with high technical readiness levels, automatic feature detection, interpretable DL, and performing simulation-based inference.