blasé

Interpretable Machine Learning for high-resolution astronomical spectroscopy.

Handles stellar and telluric lines simultaneously

We can combine stellar, telluric, and instrumental models into a unified forward model of your entire high-bandwidth, high-resolution spectrum. We can obtain best-in-class models of Earth's atmosphere, line-by-line, automatically, for free (or cheap).

Massively scalable

By using autodiff, we can fit over 10,000 spectral lines simultaneously. This enormous amount of flexibility is unavailable in conventional frameworks that do not have autodiff.

^ We do this for 10,000 lines simultaneously.

Rooted in physics

We first clone a precomputed synthetic spectrum, such as PHOENIX, and then transfer learn with data. By regularizing to the cloned model, we get the best of both worlds: data driven when the Signal-to-Noise ratio is high, and model-driven when we lack data to say otherwise.

Blazing fast with GPUs

We achieve $>60 \times$ speedups with NVIDIA GPUs, so training takes minutes instead of hours.

Get started

Visit our step-by-step tutorials or installation pages to get started. We also have deep dives, or you can read the paper. Have a question or a research project in mind? Open an Issue or email gully.

Copyright 2020, 2021, 2022, 2023 The Authors