Enhancing Stellar Distance Estimation with Uncertainty-Weighted Loss Functions in Deep Learning Models,

Accurate estimation of stellar parameters is crucial for advancing our understanding of stellar evolution and galactic dynamics. Traditional deep learning approaches for stellar parameter estimation often treat all data points uniformly, neglecting the inherent uncertainties in geometric, spectroscopic, and photometric measurements. In this study, we propose a novel approach that incorporates uncertainty-weighted loss functions to train deep learning models for stellar distance estimation. By weighting the contribution of each training sample based on its associated uncertainty, our method enables the utilization of even erroneous data, thereby facilitating the concurrent enhancement of the precision in stellar parameters estimation and the generalization of the model. We evaluate our approach using benchmark stellar datasets, demonstrating improved predictive accuracy and robustness compared to conventional loss functions.