Cosmic Collaboratory
Want to submit a model? Get started here
About the Collaboratory
The Cosmic Collaboratory is an effort to establish a community of practitioners in the niche of Machine Learning (ML) for Space Domain Awareness (SDA). While some models are available as completely open-source software, others have more restricted access (e.g. USG-only), and may not be suited for an open-source platform like Hugging Face or TorchHub. In 2024, we presented our AMOS paper “Backbone Architectures for Space Domain Awareness”, and received requests for the models described therein; however, due to access control requirements, they could not be readily shared. The Cosmic Collaboratory aims to ameliorate this by providing a more secure way to share, access, and discuss models for the SDA ML community. In the spirit of a rising tide raises all ships, we hope to learn together and enable researchers in the SDA field to rapidly design robust solutions.
Download the “Backbone Architectures” paper : https://amostech.com/TechnicalPapers/2024/Machine-Learning-for-SDA/Merry.pdf
Featured Models
- XCiT IQR Low SNR
XCiT (Cross-Covariance Image Transformer) trained on Low SNR data with IQR rolloff.
Access the model and artifacts in Inteldocs
- Swin IQR Low SNR
Shifted Window Transformer trained on low SNR data with IQR rolloff.
Access the model and artifacts in Inteldocs
- ResNet IQR Low SNR
Residual Network trained on low SNR data with IQR rolloff.
Access the model and artifacts in Inteldocs
- U-Net IQR Low SNR
High resolution U-Net trained on low SNR data with IQR rolloff.
Access the model and artifacts in Inteldocs