2025 workshop program
Augmenting Spectra Emulation with Deep Evidential Regression for the Purposes of Biomarkers Retrieval
The work
Abstract
The Biomarkers retrieval problem is to determine if Carbon-based macromolecules, typically indicative of life, are present in exoplanet atmospheres. The James Webb Space Telescope is offering an unprecedented view into these exo-worlds with the 30–100 milli arc second resolution Near Infrared Camera (NIRCam) in tandem with the Near-Infrared Spectrograph (NIRSpec). With these instruments, we can now study if our measured transmission spectra contain contributions from Biomarkers such as Carbon and Ammonia. More generally though, this analysis requires extensive databases of chemical absorption spectra so that we may best probe what compounds are contributing to the measured signal. These databases are limited by the expenses of lab based measurements and DFT calculations. In this work, we present a machine learning model that can emulate chemical spectra, allowing us to produce spectral lines for unseen molecules. Our emulator also produces uncertainty estimates through deep evidential regression, which enables us to better solve the Biomarkers Retrieval problem through Hamiltonian Monte Carlo.