English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep learning algorithms are growing in popularity in the field ofexoplanetary science due to their ability to model highly non-linear relationsand solve interesting problems in a data-driven manner. Several works haveattempted to perform fast retrievals of atmospheric parameters with the use ofmachine learning algorithms like deep neural networks (DNNs). Yet, despitetheir high predictive power, DNNs are also infamous for being 'black boxes'. Itis their apparent lack of explainability that makes the astrophysics communityreluctant to adopt them. What are their predictions based on? How confidentshould we be in them? When are they wrong and how wrong can they be? In thiswork, we present a number of general evaluation methodologies that can beapplied to any trained model and answer questions like these. In particular, wetrain three different popular DNN architectures to retrieve atmosphericparameters from exoplanet spectra and show that all three achieve goodpredictive performance. We then present an extensive analysis of thepredictions of DNNs, which can inform us - among other things - of thecredibility limits for atmospheric parameters for a given instrument and model.Finally, we perform a perturbation-based sensitivity analysis to identify towhich features of the spectrum the outcome of the retrieval is most sensitive.We conclude that for different molecules, the wavelength ranges to which theDNN's predictions are most sensitive, indeed coincide with their characteristicabsorption regions. The methodologies presented in this work help to improvethe evaluation of DNNs and to grant interpretability to their predictions.

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