Conveners
Innovative ways of data analysis, data handling, machine learning, and big data
- Ioannis Kontogiannis (Leibniz Institute for Astrophysics Potsdam (AIP) Germant)
Innovative ways of data analysis, data handling, machine learning, and big data
- Meetu Verma (Leibniz-Institut für Astrophysik Potsdam (AIP))
Contemporary solar physics deals with increasing volumes of high-dimensional data from observations and realistic theoretical models alike. Machine Learning (ML) is rapidly integrated into solar and heliophysics research, facilitating dimensionality reduction, visualization and analysis. We developed a pipeline, which employed carefully selected unsupervised ML algorithms for classification...
Many solar spectral lines have poorly determined atomic parameters, such as the transition probability or the central wavelength, especially in the near-ultraviolet (NUV) range. We present a novel approach to exploit high-angular-resolution spectropolarimetric observations of the Sun to obtain precise atomic parameters of spectral lines, required for the proper analysis of solar and stellar...
Solar spectropolarimetry is entering the realm of big data. Current and future telescopes will produce data at a rate that will make it hard to store in a single machine and even harder to operate on the data. Thankfully, in the last decade, machine learning has experienced an enormous advance, thanks to the open possibility of training very deep and complex neural networks. In this...
A solar magnetic flux tube, like any physical system, is capable of supporting a variety of resonances, which constitute its natural response to external forcing. Apart from a few single low order eigenmodes in small scale magnetic structures, several simultaneous resonant modes were not found until recently. These eigenmodes are associated to small but coherent oscillations and require the...
Physics-informed neural networks (PINNs) provide a novel approach for numerical simulations, tackling challenges of discretization and enabling seamless integration of noisy data and physical models (e.g., partial differential equations). In this presentation, we highlight the new opportunities for solar magnetic field modelling that are enabled through physics-informed machine learning.
We...
We present results of generative deep learning as applied to a large database of solar images and discuss challenges in training and validation, in particular with distributed training on GPU supercomputers. Our dataset is based on SDO EUV data with high resolution and with excellent coverage since 2010. This dataset is thus well suited to study the application of advanced machine learning...
Observations of plasma motions in the low corona are often limited to magnetic field lines originating in active regions, which are ideal for spatial domain enhancements across individual extreme ultraviolet (EUV) images to see loops, flares, and other bright activity contrasted against dim background features.
The quiet Sun is essentially all dim background features, which requires...
Although applicable to any 3D datacube, this software was inspired by a single thought: what would the Sun look like if we could only see light from one spectral line? In an absorption line, with low emission in the central wavelengths, and high wings on either side, the blue and red cone receptors of our thought experiment would be triggered, making the Sun appear purple. For a strong, narrow...
Our knowledge of the lower solar atmosphere is mainly obtained from spectropolarimetric observations, which are often carried out in the red or infrared spectral range and almost always cover only a single or a few spectral lines. Here we compare the quality of Stokes inversions of only a few spectral lines with many-line inversions. In connection with this, we have also investigated the...
