The optical and infrared are connected @Bristol

Galaxies are traditionally modeled as composites of separable components, each dominating distinct wavelength regimes. However, this assumption fails to capture the intricate correlations between physical processes. We present a data-driven approach that leverages a neural spectral summarizer to extract compressed representations of optical Sloan Digital Sky Survey (SDSS) spectra, enabling precise predictions of infrared (IR) WISE photometry. Our model achieves near-perfect agreement with observed WISE fluxes and accurately constrains key IR-derived properties, including AGN bolometric luminosities and dust parameters. In contrast, standard SED-fitting methods struggle to achieve similar predictive power. By analyzing our neural summarizer’s feature importance, we identify critical spectral lines — CaII, SrII, FeI, [OII], and H-alpha — that drive the IR predictions, revealing the complex chronology of star formation and chemical enrichment of each galaxy. Our findings challenge conventional SED models and underscore the power of machine learning in uncovering previously overlooked astrophysical relationships.