Astro-MoE: Mixture of Experts for Multiband Astronomical Time Series

TL;DR

Astro-MoE introduces a Mixture of Experts transformer architecture tailored for multiband astronomical time series, effectively capturing heterogeneous variability patterns and improving modeling of complex astronomical signals.

Contribution

The paper presents Astro-MoE, a novel transformer model with a Mixture of Experts module designed specifically for multiband astronomical data, addressing limitations of shared-parameter models.

Findings

Effective modeling of multiband variability patterns.

Improved performance on simulated and real datasets.

Demonstrates flexibility of Mixture of Experts in astronomy.

Abstract

Multiband astronomical time series exhibit heterogeneous variability patterns, sampling cadences, and signal characteristics across bands. Standard transformers apply shared parameters to all bands, potentially limiting their ability to model this rich structure. In this work, we introduce Astro-MoE, a foundational transformer architecture that enables dynamic processing via a Mixture of Experts module. We validate our model on both simulated (ELAsTiCC-1) and real-world datasets (Pan-STARRS1).