Just to self advertise my older post. You may want to start from there ^^
I would like to use ML to work out the output (could it be a simulated collision or a real one, but it is better to start with simulated stuff since data are not publicly released yet and only experimenters have access to them) and try to reconstruct what happened.
Bookmarked for weekend reading :). Btw a lil off-tangent, have you seen Numerai? It's a pretty interesting company, sometihng to do with machine learning and data encryption for worldwide collaboration.
ML to work on pattern-recognizing the output right? If that's the case then i pretty much understood what you're trying to say..