CNN4MAGIC

This is the repository of the code used for my Master Thesis "MAGIC Deep Learning: a Performance Evaluation for Very High Energy Gamma-Ray Astrophysics", where I apply convolutional neural networks for the full analyisis of IACT MAGIC telescopes.

Topics covered

The repository solves the problems of:

1. Energy Reconstruction
2. Direction Reconstruction Using Regression techniques (Neural networks compiled to minimize the MSE loss)
3. Gamma/Hadron separation using Classification techniques (Neural networks compiled to minimize the binary cross entropy loss)

The new developed pipeline is then applied to real acquisition of the Crab Nebula, on which integral sensitivity above 150GeV is 1.12%