Robustsed

Robustsed is a term used in the context of spectral energy distribution analysis to denote a class of methods and tools designed to fit model spectral energy distributions to multiwavelength data in a way that is resistant to outliers, non-detections, and systematic errors. The aim is to produce more reliable inferences of astrophysical properties such as stellar mass, star formation rate, and dust attenuation, especially when data come from heterogeneous surveys with varying depths and noise characteristics.

Approaches rely on robust loss functions (Huber, Tukey), L1/L2 regularization, Bayesian hierarchical models, and survival analysis for upper limits; they can be implemented within maximum-likelihood or Bayesian frameworks. They often integrate upper limits, non-detections, and handling of missing data. They may also incorporate priors on physical properties and model libraries.

Applications include the fitting of galaxy spectral energy distributions from photometric surveys, estimation of photometric redshifts, and inference of properties of active galactic nuclei and star-forming regions. Benefits include reduced sensitivity to outliers, better handling of censored data, and more robust uncertainty estimates; limitations include increased computational cost, the need for careful tuning of hyperparameters, and potential bias if the chosen model or loss function is mis-specified.

See also: spectral energy distribution fitting, robust statistics, Bayesian inference, and survival analysis.