My MPhys project exploring extensions to the Standard Model of particle physics. This project involves deep learning with large datasets from the LHC.
Headline result so far: my CLs predicting model can correctly remove 45% of invalid models whilst removing only 1% of valid models. This translates to saving more than a month of processing by the LHC computing cluster out of a year total, for the recent EWKino scan.
- Create model that predicts DM relic density better than random sampling of target range
- Reduce MAE to below 25% of mean absolute target error
- Create plot of MAE in bins of true relic density
- Plot upper cutoff of relic density against number of valid but excluded pMSSM models
- Apply EWKino model to Bino-DM and evaluate performance
- Apply same relic density cutoff technique to CLs
- Characterise those models which my CLs predictor incorrectly rejects
- Recreate plots from pMSSM paper with predicted CLs