Speaker
Description
Comprehensive analysis of time-resolved spectral energy distributions (SEDs) of blazars is essential for understanding their underlying physical processes. However, when numerical kinetic models are combined with rigorous statistical inference-such as full posterior sampling instead of local optimization-the task of fitting large numbers of SEDs becomes computationally prohibitive. To address this challenge, we developed a surrogate-modeling framework based on convolutional neural networks. Trained on state-of-the-art time-dependent numerical simulations, these models accurately reproduce blazar spectra while reducing computational costs by several orders of magnitude, enabling the analysis of hundreds of SEDs. Using OJ 287, PKS 2155-304, and 1ES 1959+650 as case studies, I will demonstrate how time-resolved SED fitting uncovers systematic variations in jet microphysical and dynamical properties across different emission states.
