Speaker
Description
Forecasting high-energy flares in blazars—active galactic nuclei with relativistic plasma jets oriented toward Earth—over extended temporal horizons presents a significant challenge due to the complex variability inherent in their light curves. In this study, we investigate the long-term predictability of flare activity using over 15 years of photon flux observations from the Fermi-LAT telescope. Accurate forecasting of such flares can provide valuable insights into the physical mechanisms governing relativistic jet formation in the vicinity of supermassive binary black holes. We explore three complementary approaches to long-horizon forecasting. First, we apply time-delay embedding to capture latent temporal dynamics and forecast future flux values. Second, we employ a convolutional neural network (CNN) with explanation-guided regularization, which promotes temporal consistency and enhances model interpretability. Third, we formulate flare prediction as a binary classification task, using statistically derived thresholds to label future events. Each approach offers unique insights into the flare dynamics of blazars, and the results highlight the feasibility of early flare detection from historical flux dynamics and suggest that incorporating domain-informed regularization and interpretable deep learning offers a viable path toward reliable flare forecasting in astrophysical time series.
