A neural network has been developed and applied on helioseismic far-side images, and substantially improved the number of far-side active region detections with higher true positive rate.
An emulation of the VFISV Stokes Inversion that trains a deep
network (U-Net) to map directly from IQUV polarized light to Milne-Eddington magnetic field parameters. The accuracy of this method suggests that it could serve as a warm-start for VFISV or as a pre-disambiguation stand-in.
An unsupervised machine-learning algorithm is used on selected features derived from the polarity inversion lines (PIL) mask and difference PIL mask. It is found these features are effective in predicting flaring occurrences.
A deep learning code is trained using the Sun’s front-side observations, HMI’s magnetograms and AIA’s 304Å EUV images, to establish a relation between magnetic field and EUV flux. Then the code is applied on the STEREO/EUVI 304Å data to obtain the Sun’s far-side magnetic field.
A deep learning code is developed to enhance HMI continuum intensity images and line-of-sight magnetic field for a better spatial resolution.
A set of parameters that characterize the complexity and energy potential of solar active-regions is fed through several Machine Learning and conventional statistics algorithms to forecast solar flares.
A deep-learning method, Convolutional Neural Network, is developed to use the HMI’s line-of-sight magnetic field to forecast solar flares.
We attempt to forecast M- and X-class solar flares using a support vector machine algorithm and 4 years of HMI vector magnetic field data. With the true skill statistic as the preferred metric to estimate the algorithm performances, we obtain good predictive abilities. This may be partly due to fine-tuning the algorithm for this purpose, and to a choice of 13 SHARP parameters obtained from vector magnetograms.