Using Artificial Neural Networks to Discover Exoplanets
Abstract: Our project idea is to make a neural network to detect the location and existence of exoplanets in other solar systems by using data collected on those solar systems taken by NASA probes as inputs to the neural network. This project is supposed to automate the detection and recognition of planets that are outside of our solar system. The current method of detecting exoplanets is by special patterns in data collected in the solar system the exoplanet is in, which is what neural networks are best at. The accompanying program to the neural network would remove any invalid data before feeding it into the network, as well as interpreting the output of the network. Our intention is to make the source code public and establish an API to the trained network, or release the parameters of the trained network. This model could be deployed on outer space satellites to automate the process of discovering new exoplanets.
This project implemented many features outlined in the course. Neural Networks play a big role in this project as the discovery of exoplanets is hugely based on mathematics, and neural networks excel at these complicated maths. Also, classification plays a big role because many astrophysicists use live data from telescopes such as Kepler to classify different galactic bodies. Preprocessing tools such as regularization and standardization would be used due to the myriad of data types and the number of features for better performance of the model.