Decorrelated classifiers for robust signal detection at the Large Hadron Collider via Optimal Transport

Oregon Convention Center 

New physics searches are usually done by training a supervised classifier to separate a signal model from a background model (known physics Standard Model). However, even when the signal model is correct, systematic errors in the background model can influence supervised classifiers and might adversely affect the signal detection procedure. To tackle this problem, one approach is to find a classifier constrained to be decorrelated with one or more protected variables, e.g. the invariant mass. Then use this classifier to find a signal-rich region where one can perform the signal detection test using the protected variable. We perform the decorrelation by considering an optimal transport map of the classifier output that makes it independent of the invariant mass for the background. We then estimate the signal strength in the signal-rich region to detect the presence of signal, using the experimental data on the invariant mass. We compare and contrast this decorrelation method with previous approaches, show that the decorrelation procedure is robust to background misspecification, and analyze the power of the test.