Skip to content

Generative and Interpretable Deep Learning for Stellar Spectra

License

Notifications You must be signed in to change notification settings

teaghan/Cycle_SN

Repository files navigation

Cycle-StarNet

Dependencies

-PyTorch: pip install torch torchvision

-h5py: pip install h5py

-scikit-learn: pip install -U scikit-learn

Generative and Interpretable Deep Learning for Stellar Spectra

This project aims to bridge the gap between two different sets of stellar spectra. Although the underlying physics that produces the two sets may be the same, the data can visually appear very different for a variety of reasons. Cycle-StarNet is meant to determine the commonalities between two sets of spectra (ie. the physical parameters) and learn how to transform one set of data into the other.

Figure 1: Overview of the proposed method (top) and a simplified diagram of the Cycle-StarNet Architecture (bottom).

Figure 2: Examples of two spectra (taken from our New Lines Project) from opposite domains that have the same stellar parameters. When mapping the synthetic spectrum (which has an incomplete line list) to the observed domain, the resulting spectrum is a much better fit to the observed spectrum and the missing information is now present in the transferred spectrum.

Getting Started

For an in depth explanation on the method and the projects shown here, please checkout our paper.

To get started with our application of Domain Adaption with real observed spectra, take a look here.

To get started with our New Lines Project, take a look here.

Citing this work

Here's the BibTeX:

@article{OBriain_2021,
	doi = {10.3847/1538-4357/abca96},
	url = {https://doi.org/10.3847/1538-4357/abca96},
	year = 2021,
	month = {jan},
	publisher = {American Astronomical Society},
	volume = {906},
	number = {2},
	pages = {130},
	author = {Teaghan O'Briain and Yuan-Sen Ting and S{\'{e}}bastien Fabbro and Kwang M. Yi and Kim Venn and Spencer Bialek},
	title = {Cycle-{StarNet}: Bridging the Gap between Theory and Data by Leveraging Large Data Sets},
	journal = {The Astrophysical Journal}
}