ABCDisCo: Automating the ABCD Method with Machine Learning

Abstract

The ABCD method is one of the most widely used data-driven background estimation techniques in high energy physics. Cuts on two statistically-independent classifiers separate signal and background into four regions, so that background in the signal region can be estimated simply using the other three control regions. Typically, the independent classifiers are chosen "by hand" to be intuitive and physically motivated variables. Here, we explore the possibility of automating the design of one or both of these classifiers using machine learning. We show how to use state-of-the-art decorrelation methods to construct powerful yet independent discriminators. Along the way, we uncover a previously unappreciated aspect of the ABCD method: its accuracy hinges on having low signal contamination in control regions not just overall, but relative to the signal fraction in the signal region. We demonstrate the method with three examples: a simple model consisting of three-dimensional Gaussians; boosted hadronic top jet tagging; and a recasted search for paired dijet resonances. In all cases, automating the ABCD method with machine learning significantly improves performance in terms of ABCD closure, background rejection and signal contamination.

Anomaly Detection for Physics Analysis and Less than Supervised Learning

“…In both approaches, the data are used directly to transport predictions from a control region to the signal sensitive region instead of relying on simulation for this extrapolation. In the ABCD method, two classifiers f and g and two working points a and b are constructed and then four regions called A, B, C and D are defined by f ≶ a and g ≶ b (for a machine learning version of ABCD, see [21][22][23]). If f and g are independent, then one can relate the background prediction in the region f > a and g > b to the other three regions.…”

“…One can simply remove mass-sensitive features from the training, but powerful classifiers can learn the mass indirectly through subtle correlations with other useful features. A variety of decorrelation techniques exist to solve this problem [21,[26][27][28][29][30][31][32][33][34][35][36][37][38][39]. In the context of neural networks, one can add terms to the loss function to achieve automatic decorrelation:…”

Simulation-Assisted Decorrelation for Resonant Anomaly Detection

“…A key challenge facing such methods is that the machine learning classifiers must be relatively independent from the resonant feature, for otherwise artificial bumps can be formed. Many automated decorrelation methods have been proposed to ensure that classifiers are decorrelated from particular features by construction [39][40][41][42][43][44][45][46][47][48][49][50], but they may not apply in all cases. In particular, weakly supervised approaches that learn directly on the signal region cannot be simply combined with a decorrelation scheme because such an approach could degrade the performance in the presence of a signal.…”

Bayesian Probabilistic Modelling for Four-Tops at the LHC

“…In particular, the advent of unsupervised and weaklysupervised Machine Learning (ML) techniques has allowed for the development of broad model independent NP search and characterisation strategies [1]. Simultaneously, there have been important efforts to reduce reliance of LHC measurements on Monte Carlo (MC) simulations of hadronic processes [2][3][4][5][6].…”

“…So the model would over-parameterize the data making the inclusion of mixtures redundant. 2 Therefore the key insight is to instead write down a mixture model in terms of p(j n |z n ) and p(b n |z n ), such that the correlations between N j and N b in the dataset are parameterized by the class label alone. The number of parameters in this model is 2 × (d j + d b − 2) + 1.…”