Btw referencing your earlier reply -

To say the truth, I would like to use this to develop new techniques for looking for new phenomena at particle colliders.

So machine-learning will be applied to which part of this process? Pattern recognition for collider configurations, or the output (both "live" and archived data) ?

In a collisions at the LHC, we get final state products that consist of many particles. We then cluster them in some way (that contains some free parameters and methods) and study the output of this clustering. From this output, we can study several observables, properties, etc...

I think this will need some lecture of its own - I'll go do some research on it. Hopefully there's something decent online :)