SoylentNews is people
SoylentNews
Develop Code to Study the Higgs Boson and Win Cash Prizes.
Physicists at the Large Hadron Collider would like your help in studying the Higgs Boson. It' s true that they found the Higgs boson-which is responsible for giving all particles their mass-nearly two years ago, but its exact behavior is still mysterious. Now, the scientists are asking coders to develop algorithms that can reveal the Higgs' properties. The Higgs Boson Machine Learning Challenge will reward successful coders with up to $ 7,000 in actual money. But it' s hard to put a price on the chance to be involved in one of the biggest scientific discoveries of the decade. Interested? Here' s the situation. At the LHC, protons are smashed together at colossal energies, creating a chaotic shower of particles. Physicists have to hunt through the noisy mess of other particles to see the Higgs' weak decay signal. They already have code that can pull out the Higgs signal from this noise (in fact, researchers at the Higgs-hunting ATLAS experiment didn't actually see the enigmatic particle in their detector, simply its decay signal). But the scientists think the public might be able to help them get a sharper signal and figure out what the Higgs is really like.
The contest started up about a week ago and already has nearly 200 participants. But the final prizes won't be awarded until September, so there's plenty of time for interested budding scientists to get involved.
(Score: 2) by visaris on Wednesday May 21 2014, @10:08PM
> "No knowledge of particle physics is required."
> Well, this should be entertaining.
This is actually fairly accurate in many cases of the application of machine learning. The whole idea is to get the machine to find the patterns between the inputs (the features) and the output (the classification: "tau tau decay of a Higgs boson" versus "background"). So real-valued inputs map to 0 or 1. One doesn't need to have intimate knowledge (though, that may help things, and does in practice) of physics. One just needs to be good at the abstract problem of mapping some set of reals -- likely some combination of a subset of the given features and of some function(s) of the given features) -- to the set {0,1}. The "magic" will be in the selection of these features (and functions of features) in combination with the back-end learning mechanism, such as a SVM, NN, ensemble, etc.