Machine Learning and Optimization for Collision Selection in High
Energy Physics The field of high energy physics aims to probe fundamental mysteries about the nature of matter and the laws that govern its interactions. The primary experimental tools for doing so are particle accelerators like the FermiLab Tevatron near Chicago and the Large Hadron Collider currently under construction near Geneva. These enormous machines accelerate and annihilate protons and anti- protons so as to generate exotic particles that have not existed naturally since the first moments after the Big Bang. Filtering and analyzing the data resulting from these accelerators poses great computational challenges. The goal of this project is to address those challenges using cutting edge stochastic optimization and machine learning methods. In particular, this project aims to improve the precision of mass measurements of the top quark, the largest observed subatomic particle, from data collected on the Tevatron. Doing so requires improving the process of collision selection, wherein collisions resulting in top quarks are separated from those resulting in other particles. Previous results have already demonstrated that stochastic optimization techniques can significantly improve collision selection, yielding more precise mass measurements that would otherwise require literally millions of dollars and hundreds of person-years to generate. However, much work remains to be done to improve this process and rigorously compare performance among different optimization and supervised learning techniques.
The resulting methodology will then be applied to collision selection
problems resulting from new data generated by the LHC, with the goal
of aiding the search for the Higgs boson, a critically important but
as yet unobserved particle theorized to give mass to all other
particles through its interactions.
|
|
Keywords: |
High Energy Physics, Stochastic Optimization, Supervised Learning |
Study: |
Artificial Intelligence |
Contact: |
Shimon Whiteson |
Location: |
Universiteit van Amsterdam |
References: |
http://staff.science.uva.nl/~whiteson/pubs/b2hd-whitesoniaai07.html |