Inspiration

The idea of being able to look out to the stars to find answers to problems here on Earth has always been an incredible influence on the lives of scientists, and to be able to work with these type of datasets as students is a privilege.

What it does

This project is aimed at detecting anomalies in image based datasets; particularly in astronomical data.

How I built it

We followed a paper from Siqi Wang on "Effective End-to-end Unsupervised Outlier Detection via Inlier Priority of Discriminative Network" and ran this code on the Galaxy10 dataset provided on the astroNN documentation. https://papers.nips.cc/paper/2019/file/6c4bb406b3e7cd5447f7a76fd7008806-Paper.pdf

Challenges I ran into

Running a machine learning algorithm in one day is not an easy task, and many papers on the topic are supervised or semi-supervised neural networks which is not accurate to our needs. Having to outfit this challenging paper to our needs was the main struggle we faced.

What I learned

We learned a lot about astrophysical datasets, how to access them, and how scientists are using them. As well, we learned methods for anomaly detection using machine learning which none of us knew beforehand.

What's next for Astro Anomalies

Graduate.

Built With

Share this project:

Updates