Multi-Variable Stellar Parameter Estimation Using Residual Multitask Neural Networks
Researchers have developed a new neural network architecture for estimating stellar parameters from astronomical spectra. This end-to-end pipeline utilizes a fully connected multitask neural network with residual blocks, optimized through Bayesian methods. The model preprocesses spectra by standardizing them, normalizing target variables like effective temperature, metallicity, and surface gravity, and augmenting data with Gaussian noise. It achieved competitive accuracy with significantly lower complexity compared to deeper baseline models. AI