**➥ Tip!** Refine or expand your search. Authors are sometimes listed as 'Smith, J. K.' instead of 'Smith, John' so it is useful to search for last names only. Note this is currently a simple phrase search.

[ascl:2003.010]
MARGE: Machine learning Algorithm for Radiative transfer of Generated Exoplanets

MARGE (Machine learning Algorithm for Radiative transfer of Generated Exoplanets) generates exoplanet spectra across a defined parameter space, processes the output, and trains, validates, and tests machine learning models as a fast approximation to radiative transfer. It uses BART (ascl:1608.004) for spectra generation and modifies BART’s Bayesian sampler (MC^{3}, ascl:1610.013) with a random uniform sampler to propose models within a defined parameter space. More generally, MARGE provides a framework for training neural network models to approximate a forward, deterministic process.

[ascl:2003.011]
HOMER: A Bayesian inverse modeling code

HOMER (Helper Of My Eternal Retrievals) is a machine-learning-accelerated Bayesian inverse modeling code. Given some data and uncertainties, the code determines the posterior distribution of a model. HOMER uses MC^{3} (ascl:1610.013) for its MCMC; its forward model is a neural network surrogate model trained by MARGE (ascl:2003.010). The code produces plots of the 1D marginalized posteriors, 2D pairwise posteriors, and parameter history traces, and can also overplot the 1D and 2D posteriors for comparison with another posterior. HOMER computes the Bhattacharyya coefficient to compare the similarity of two 1D marginalized posteriors.

[ascl:2307.055]
plan-net: Bayesian neural networks for exoplanetary atmospheric retrieval

Cobb, Adam D.; Himes, Michael D.; Soboczenski, Frank; Zorzan, Simone; O'Beirne, Molly D.; Güneş Baydin, Atılım; Gal, Yarin; Domagal-Goldman, Shawn D.; Arney, Giada N.; Angerhausen, Daniel

plan-net uses machine learning with an ensemble of Bayesian neural networks for atmospheric retrieval; this approach yields greater accuracy and more robust uncertainties than a single model. A new loss function for BNNs learns correlations between the model outputs. Performance is improved by incorporating domain-specific knowledge into the machine learning models and provides additional insight by inferring the covariance of the retrieved atmospheric parameters.