If you've ever wanted to get involved with a particle physics analysis but not had the opportunity to work with one of the largest scientific collaborations in the world, your luck has changed!
The ATLAS experiment is running a machine learning challenge to try and gain input and expertise from people all over, who might have a new way of looking at our data to discover the Higgs particle.
Machine learning encompasses the branch of computer science where you use clever algorithms (such as boosted decision trees and neural networks) which learn from known data to help classify unknown data.
Here in the particle physics community, the known data falls into two categories (a binomial problem) - signal and background. The signal is the process we want to discover. The background is all the other processes which look almost identical to the process we really want to see.
If our signal was so unique and different from the background continuum then we would not need to work out ways to discriminate between the two. Sadly Nature is not that kind to us, and often if we are looking for one particle physics process to produce a final state of particles, there will always be other ways (other Feynman diagrams) which could produce the same outcome.
The challenge here has simulated data from Higgs to di-tau events as the signal. From a statistical standpoint, if you can correctly categorise the signal and background into their respective categories then you can evaluate the discovery sensitivity or discovery significance. Information is provided on calculating this, and this will be the marker with which to rank the entrants.
People wishing to participate have until the 15th September 2014 to submit an application, and you can submit a limited number per day. Further information about the machine learning tools which are freely available is provided on the webpages and the tutorial pages.
If you are good enough you could win up to $7000. Not a bad prize for doing some coding. Already there have been over 800 entrants and the great thing about this is that you don't need to understand the physics involved to manipulate the data, extract information from the variables, and combine it all together to correctly classify data. The only benefit particle physicists get is knowing what kind of correlations to expect between these variables, but in fact, a perfect machine learning algorithm should be able to find and exploit these correlations anyway.
Additional Links
CERN Blog Post Extra Information From LAL
