Dr Adrian Bevan Project Abstracts

Dr Adrian Bevan Project Abstracts

BSc Projects

Learning to discover: the Higgs boson machine learning challenge 
In 2012 particle physicists discovered a new particle, thought to be the long sought after Higgs boson.  As a part of the CERN Open data programme a challenge was set in 2015 for people to try and improve on the Higgs boson selection using multivariate techniques.  A group of non-particle physicists managed to obtain a better performance than the particle physicists and won the challenge. This project starts with a brief literature review in terms of the relevant machine learning and particle physics before embarking on the machine learning challenge. Requires as strong background in C++, Requires MT1, MT2, MT3 and students are expected to take SDA if they opt for this project.

Deep Learning applied to LHC analysis
Modern machine learning algorithms are focusing on deep networks, including convolutional neural nets to learn from features ranging from multidimensional images to high dimensional feature spaces derived from physics quantities.  These techniques can provide additional discriminating power between signal and background that may lead to improved precision on physics outputs.  Requires as strong background in C++, Requires MT1, MT2, MT3 and students are expected to take SDA if they opt for this project.

Build a radiation detector
The basic unit of a large scale silicon tracker is a reversed biased diode. Complex arrays of these are used to build precision digital "cameras" for projects like the Large Hadron Collider. This project invovles studying diodes fabricated by Micron Semiconductor Ltd. and exploring how to turn the processed silicon sensors into detectors that can be read out, tested in the lab and exposed to radioactive sources.. 

MSci (Review) Projects

CP violation and the CKM Matrix (Review only)
In 1972 Kobayashi and Maskawa proposed a 3x3 quark mixing matrix to account for CP violation observed in the decay of neutral kaons. In 2008 having had their model verified by the B Factory experiments (BaBar and Belle), they shared a Nobel Prize in physics. The amount of CP violation that we know about is a billion times too small to explain the matter-anti matter asymmetry in the universe, and the CKM matrix has only been "precisely" tested with s and b quark flavor transitions. The precision of these tests is of the order of 10%, which leaves a lot of scope for physics beyond the standard model to affect what we measure. The s and b quarks are both down type quarks, and a programme of measurements using c quark transitions to test the CKM matrix has recently been outlined. This review will delve into the known aspects of the CKM matrix and explore possible tests to examine the validity of the model further. 

Searching for magnetic monopoles
The magnetic monopole is a hypothetical particle that necessitates quantisation of charge.  A number of monopole variants exist proposed by Dirac and subsequently other theorists.  This project will review the different types of monopole and experimental constraints placed on the existence of such a particle [review project], or work on searching for monopoles using data from the MoEDAL experiment. 

Multivariate analysis and optimisation algorithms (Review option and full project option)
Multivariate analysis techniques are used in a number of scientific applications, including the search for the Higgs boson decays to specific final states and rare decays. This review will focus on modern multivariate algorithms such as support vector machines to compare and contrast with some of the more widely known approaches.

  • The full project option requires MT1, MT2, MT3, SDA and C++ and starts with a review of methods before applying this knowledge to the use of the ROOT based package TMVA in the context of an LHC inspired problem.
  • The review option requires MT1, MT2 and MT3 and students are expected to take SDA.

Build a radiation detector
The basic unit of a large scale silicon tracker is a reversed biased diode. Complex arrays of these are used to build precision digital "cameras" for projects like the Large Hadron Collider. This project invovles studying diodes fabricated by Micron Semiconductor Ltd. and exploring how to turn the processed silicon sensors into detectors that can be read out, tested in the lab and exposed to radioactive sources.. 

Learning to discover: the Higgs boson machine learning challenge 
In 2012 particle physicists discovered a new particle, thought to be the long sought after Higgs boson.  As a part of the CERN Open data programme a challenge was set in 2015 for people to try and improve on the Higgs boson selection using multivariate techniques.  A group of non-particle physicists managed to obtain a better performance than the particle physicists and won the challenge. This project starts with a brief literature review in terms of the relevant machine learning and particle physics before embarking on the machine learning challenge. Requires as strong background in C++, Requires MT1, MT2, MT3 and SDA.

Deep Learning applied to LHC analysis
Modern machine learning algorithms are focusing on deep networks, including convolutional neural nets to learn from features ranging from multidimensional images to high dimensional feature spaces derived from physics quantities.  These techniques can provide additional discriminating power between signal and background that may lead to improved precision on physics outputs.  Requires as strong background in C++, Requires MT1, MT2, MT3 and SDA.

Juno Champion

The school holds Juno Champion status, the highest award of this IoP scheme to recognise and reward departments that can demonstrate they have taken action to address the under-representation of women in university physics and to encourage better practice for both women and men.