Machine learning in the hunt for the Higgs boson
Inria is a partner in the HiggsML automatic learning challenge which aims to identify signals from this particle in the results from high energy collisions generated at CERN.
The discovery of the Higgs boson first hit the headlines two years ago. At long last, the mass of this elementary particle had been determined and in the ensuing excitement, two of the physicists involved, François Englert and Peter Higgs himself, were awarded the 2013 Nobel Prize for Physics. Their colleague Robert Brout was died en 2011.
The challenge was now to measure and understand the characteristics of the Higgs boson and to determine the extent to which these characteristics fitted into the standard model of particle physics.
The Atlas experiment on the Large Hadron Collider (LHC) at CERN was used to look for new particles generated when two beams of high-energy protons collide head on. Recent observations at Atlas have revealed a signal from the Higgs boson as it disintegrates into a pair of tau particles. This behaviour is an important step in gaining a fuller understanding of the Higgs boson, as the number of disintegrations into other particles is an essential property of all particles.
However, the signal from the disintegration is tiny in comparison with the background noise. The problem of separating this signal from the noise lies behind the launch of the Higgs boson machine learning challenge by the Atlas experiment and INRIA. Machine learning is a scientific discipline concerned with the development, analysis and implementation of potentially automatable methods to enable a ‘machine’ (in the widest sense of the word) to evolve through a learning process and carry out tasks that are difficult or impossible to perform using traditional algorithmic techniques. This is the first time that the Atlas experiment has released part of the simulated data used by its physicists to optimise their analyses.
Looking for tau particles
The aim of the challenge is to explore the potential of the latest machine learning techniques as a way improving the analysis of the data generated in the Atlas experiment. It also seeks to promote closer collaboration between high-energy physicists and experts in the field of machine learning software.
No knowledge of particle physics is needed in order to take part in the challenge. Using simulated data having the same characteristics as real events detected by Atlas, the challenge is to classify a series of events as originating from either a disintegration of a Higgs boson into two tau particles, or background noise.
The challenge runs from mid-May to September 2014. Three prizes of several thousand US dollars are being offered. In addition, the developers of the most interesting solutions will be invited to CERN to discuss their results with the high-energy physicists involved. The challenge represents a unique opportunity for machines to help develop human understanding of the Universe.