Treating the measurement of the same-sign W polarization fraction as a class imbalance problem. This is one of the two cases studies in my paper arXiv:1905.00339.
A couple of papers, arXiv:1510.01691, and more recently arXiv:1812.07591, have used deep learning to determine the polarization fraction, WL WL / Σi, j Wi Wj, in same-sign WW scattering.
In this reaction two protons (p) collide at the Large Hadron Collider and produce two jets (j), collimated sprays of hadronic particles, and two W bosons with the same electric charge. This process is interesting as a probe of the unitarization (probability conservation) mechanism in the Standard Model (SM) of particle physics.
The polarization fraction is predicted to be small in the SM, ~7%. Thus there is an imbalance of events where both W's are longitudinally polarized vs. when one or none is longitudinally polarized. This motivates trying to treat this as a class imbalance problem, something which neither of the above papers do.
Clone repository
git clone https://github.com/christopher-w-murphy/Class-Imbalance-in-WW-Polarization
Install requistes if necessary. Running
pip install -r env/minimal.txt
from the main directory of this repo will install pandas, scikit-learn, imbalanced-learn, lightgbm, and , tensorflow. The full pip freeze in available in the env folder as is my Jupyter environment.