You are viewing a single comment's thread from:
RE: A crash course on particle physics (towards our steemSTEM meetup at CERN) - 5 - The challenges of the searches for new phenomena
Are neural networks used to parse through all of those needles, or at least to cut through the most ubiquitous collisions? I imagine there's pattern recognition involved that might be suited to learning algorithms.
(Upon further thought, I feel like the answer must be yes - the alternative being all of the data being parsed by human beings, which can't be the case.)
I'll pose a second question should the first one be an easy, obvious "yes" - do the searching algorithms become more efficient as more collisions are examined over time?
Beware - technical stuff...
Machine learning methods are actually used for more than 30 years in particle physics. They are not really based on image recognition, but more on the fact that some properties of the collisions are different for the signal and the background. The output then consists in a decision of the event being more signal-like or background-like.
Yes and no. You have statistical uncertainties that are reduced with the amount of collisions. But you also have a bunch of uncertainties inherent to the method and those are independent of the amount of collisions. For many searches, we are now systematics limited and not statistics limited anymore. Therefore, we really need huge improvements in the analysis techniques to get a better sensitivity to the signals.