just add a perfunctory change
A method for building data-driven models for high-resolution stellar spectra. Designed with APOGEE2, Gaia-ESO, RAVE and HERMES data in mind, but applicable generally.
- Melissa Ness (MPIA)
- David W. Hogg (NYU; MPIA)
- Hans-Walter Rix (MPIA)
Copyright 2014 the authors.
The Cannon is free software made available under the MIT License. For details see the file LICENSE.md.
Annie Jump Cannon successfully organized stellar spectra into the OBAFGKM sequence without stellar models. Indeed, her data-driven understanding of stellar spectra made subsequent theoretical work possible. We don't need complete and accurate stellar models to make important new progress on all stellar astrophysics projects.
Although The Cannon is fully data driven, it does not follow the usual pattern of "machine learning". At it's core, The Cannon is a generative model (forward model) of noisy spectrographic data. For this reason, it is very forgiving in the face of substantial noise, missing and heterogeneous data, and low signal-to-noise.
- Empirical models for stellar spectra.
- Data-driven label transfer by supervised methods.
- Creation of a standard-star-based chemical abundance system.
- Cross-calibration of stellar surveys using co-observed stars.
- Probabilistic training of models with noisy labels.
This is a perfunctory edit, as per Prof. Hogg's instructions