codeforgoodconf / black_holes

This project was created at CodeForGood 2017
5 stars 0 forks source link
astrophysics codeforgood conference flask non-profit physics portland research

BlackholeNotBlackhole

This webapp enabled scientists to quickly classify different types of spectra collected by the Sloan Digital Sky Survey. Through an intuitive user interface, scientists build a training set for automatic black hole classification and can curate the results of expiremental automatic classification algorithms.

Background

Is it possible to use machine learning to reliably identify 'fossil' black holes in the provided spectra?

A 'fossil' black hole exists in a galaxy with large amounts of Helium II (He II). We can write a script to filter out graphs without He II, BUT galaxies with Wolf-Rayet (WR) stars also have He II. WR stars leave a 'bump' in the graph at a specified interval, but the bump is not well defined. There is no known way to calculate whether a graph has this WR bump or not. That's where machine learning comes in. We want to see if the WR bump can be found using a neural net. Using machine learning to find the WR bump in graphs will allow us to subtract WR bump graphs from the He II graphs. Thus we will have a list of spectra with He II and no WR stars, leaving us with spectra that have 'fossil' black holes.

For more details, please see ML_Info/Project_Information.pdf