An Ensemble of Bayesian Neural Networks for Exoplanetary Atmospheric Retrieval

Abstract

Machine learning (ML) is now used in many areas of astrophysics, from detecting exoplanets in Kepler transit signals to removing telescope systematics. Recent work demonstrated the potential of using ML algorithms for atmospheric retrieval by implementing a random forest (RF) to perform retrievals in seconds that are consistent with the traditional, computationally expensive nested-sampling retrieval method. We expand upon their approach by presenting a new ML model, plan-net, based on an ensemble of Bayesian neural networks (BNNs) that yields more accurate inferences than the RF for the same data set of synthetic transmission spectra. We demonstrate that an ensemble provides greater accuracy and more robust uncertainties than a single model. In addition to being the first to use BNNs for atmospheric retrieval, we also introduce a new loss function for BNNs that learns correlations between the model outputs. Importantly, we show that designing ML models to explicitly incorporate domain-specific knowledge both improves performance and provides additional insight by inferring the covariance of the retrieved atmospheric parameters. We apply plan-net to the Hubble Space Telescope Wide Field Camera 3 transmission spectrum for WASP-12b and retrieve an isothermal temperature and water abundance consistent with the literature. We highlight that our method is flexible and can be expanded to higher-resolution spectra and a larger number of atmospheric parameters.

Interpreting High-resolution Spectroscopy of Exoplanets using Cross-correlations and Supervised Machine Learning

“…In contrast, the BNN posterior sits in the middle of the degenerate peaks, and remains tightly constrained around the offset value. It is worth noting that this implementation of the BNN is not equivalent to the one used in Cobb et al (2019), as they use a different form of the likelihood which has not been tested on such high-resolution data. Cobb et al (2019) In Cobb et al (2019), it was suggested that the random forest in Márquez-Neila et al (2018) has the potential to produce over-confident, incorrect posteriors based on a mock retrieval from a test dataset.…”

“…It is worth noting that this implementation of the BNN is not equivalent to the one used in Cobb et al (2019), as they use a different form of the likelihood which has not been tested on such high-resolution data. Cobb et al (2019) In Cobb et al (2019), it was suggested that the random forest in Márquez-Neila et al (2018) has the potential to produce over-confident, incorrect posteriors based on a mock retrieval from a test dataset. This forest was trained on WFC3 spectra with 13 data points and predicted 5 parameterstemperature, free chemical abundances of H 2 O, HCN and NH 3 , and a grey cloud opacity, κ 0 .…”

“…This forest was trained on WFC3 spectra with 13 data points and predicted 5 parameterstemperature, free chemical abundances of H 2 O, HCN and NH 3 , and a grey cloud opacity, κ 0 . The opacities were calculated with HELIOS-K (Grimm & Heng 2015), using the The mock spectrum tested on by Cobb et al (2019) has T = 1479.6K, log X H2O = −9.79, log X HCN = −9.04, log X NH3 = −5.91, and log κ 0 = 1.87. The retrieved posterior for NH 3 was tightly constrained and offset from the correct value, which was used to infer that the forest could produce spurious results.…”

Optimizing exoplanet atmosphere retrieval using unsupervised machine-learning classification

“…In this section, we introduce our algorithm for generating informed priors in the context of transmission spectroscopy of exoplanetary atmospheres. This is an area which has already seen some application of machine learning, for example through the use of neural networks (Cobb et al 2019) or supervised machine learning algorithms (Márquez-Neila et al 2018). These methods generally look to the machine learning algorithm to do the full retrieval of atmospheric parameters from data.…”

Peeking inside the Black Box: Interpreting Deep-learning Models for Exoplanet Atmospheric Retrievals

“…On the neural network front, utilized a Generative Adversarial Network (GAN), a DL network architecture that can generate the closest synthetic spectrum and its associated atmospheric properties for a given observed spectrum. Cobb et al (2019) developed a Bayesian Neural Network to model the posterior distribution between atmospheric parameters. To speed up the computationally expensive radiative transfer simulation process, Himes et al (2020) trained an ML surrogate forward model and demonstrated its potential to significantly reduce retrieval time.…”