Winners announced for the Space Weather Competition. Visit http://esws2020.iopconfs.org/winners for details.
Facilitating the systematic application of machine-learning algorithms to solar flare and eruption forecasting: the SWAN-SF benchmark dataset
Manolis K. Georgoulis - Academy of Athens
Session: The solar sources of space weather
Machine and deep learning have seen an explosive growth in recent years and are considered important contenders to address real-world, Big-Data problems. Space-weather forecasting and its various applications are no exception, with numerous recent publications relying on machine- and deep-learning to break ground in nowcasting and forecasting problems. Progress, however, implies accounting for longstanding notions of data, model and performance verification. The latter, in particular, is of little meaning when data are not representative or fixed for different machine-learning methods. A solid path toward data verification is via benchmark datasets that enable competing methodologies to demonstrate their applicability, efficiency and effectiveness. A particularly well-curated benchmark dataset is the Space Weather Analytics for Solar Flares (SWAN-SF) that was constructed by the Georgia State University’s Data Mining Lab. SWAN-SF comprises ~4,100 multivariate timeseries of 51 metadata parameters representing solar active regions in flare-quiet, pre- and post-flaring situations. Five independent data partitions covering the entire solar cycle 24 allow easy and interchangeable use for training, testing and validation. The SWAN-SF benchmark dataset can both enhance physical insight and optimize method evaluation. The SWAN-SF is available at Harvard Dataverse (DOI: 10.7910/DVN/EBCFKM) and is discussed in a recent Nature Science Data paper (DOI: 10.1038/s41597-020-0548-x).
16 July 2020
Abstract submission opens:
16 July 2020
European Space Weather Medals:
6 September 2020
25 September 2020
Registration deadline: [extended]
10 October 2020
Abstract submission deadline:
4 September 2020