Final Release before presentation

Last Release before defending thesis

Final steps, training to be done

v2

contconv more efficient

First approach with continuous convolutions

Created an n-body simulation for galaxy collisions (?). First approach was to use continuous convolutions to predict accelerations given a single frame. Due to computational limitations, training wasn't as effective as expected. Error still builds up rapidly not giving faithful simulations.
Next steps are:

1. Using past information
2. Add some attention (somehow?)
3. Use graph neural networks