Explainable AI for ML jet taggers using expert variables and layerwise relevance propagation

• Published in: The journal of high energy physics 2021-05, Vol.2021 (5), p.1-36, Article 208
• Main Authors: Agarwal, Garvita, Hay, Lauren, Iashvili, Ia, Mannix, Benjamin, McLean, Christine, Morris, Margaret, Rappoccio, Salvatore, Schubert, Ulrich
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Artificial neural networks; Classical and Quantum Gravitation; Classification; Classifiers; Convolution; Decision making; Elementary Particles; Explainable artificial intelligence; Hadron-Hadron scattering (experiments); High energy physics; Jet substructure; Jets; Neural networks; Physics; Physics and Astronomy; Propagation; Quantum Field Theories; Quantum Field Theory; Quantum Physics; Regular Article - Experimental Physics; Relativity Theory; String Theory; Variables
• ISSN: 1029-8479; 1029-8479
• DOI: 10.1007/JHEP05(2021)208

A bstract A framework is presented to extract and understand decision-making information from a deep neural network (DNN) classifier of jet substructure tagging techniques. The general method studied is to provide expert variables that augment inputs (“eXpert AUGmented” variables, or XAUG variables), then apply layerwise relevance propagation (LRP) to networks both with and without XAUG variables. The XAUG variables are concatenated with the intermediate layers after network-specific operations (such as convolution or recurrence), and used in the final layers of the network. The results of comparing networks with and without the addition of XAUG variables show that XAUG variables can be used to interpret classifier behavior, increase discrimination ability when combined with low-level features, and in some cases capture the behavior of the classifier completely. The LRP technique can be used to find relevant information the network is using, and when combined with the XAUG variables, can be used to rank features, allowing one to find a reduced set of features that capture part of the network performance. In the studies presented, adding XAUG variables to low-level DNNs increased the efficiency of classifiers by as much as 30-40%. In addition to performance improvements, an approach to quantify numerical uncertainties in the training of these DNNs is presented.