This is the authors' abstract. We don't add key points for this paper.
The temporal morphology of flare light curves encodes the underlying flare physics, and deviations from the typical flare profile may indicate the presence of mechanisms not captured by a standard flare model. To search for such time-domain deviations from a “standard” flare, we develop an unsupervised deep-support-vector data-description model, which learns a compact representation of normal flares, against which unseen anomalous flares are identified. The model is trained on synthetic light curves with a “normal” flare morphology, generated from existing analytical flare-trend models with noise. Using the distribution of normal flare data, we introduce a probabilistic flare anomaly index (FLAI) which allows for separating flare light curves into three distinct classes: normal data (ND), weak anomalies (WAs), and strong anomalies (SAs). Application of the FLAI to the Kepler flare catalogue (white light) reveals that 36% and 30% of events belong to the WA and SA classes, respectively. For M- and X-class solar flares from the STIX flare list, 25% and 32% of events in the 15–25 keV channel are classified as WAs and SAs, respectively, versus 15% for both WA and SA classes in the 4–10 keV channel. Thus, anomalous flares appear more frequently in the STIX high-energy channel. These results show that both solar and stellar flares often deviate from the normal flare population used for model training, suggesting departures from the standard flare scenario, such as modified energy release and dissipation, or the development of wave and oscillatory processes in flare sites.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Белов et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: