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

TL;DR

This paper introduces a framework combining expert variables and layerwise relevance propagation to interpret and improve deep neural network classifiers for jet substructure tagging, enhancing understanding and performance.

Contribution

It presents a novel method integrating expert variables with LRP for interpretability and performance boost in jet tagging neural networks.

Findings

Adding XAUG variables increased classifier efficiency by up to 40%.

LRP combined with XAUG variables helps identify key features and interpret classifier decisions.

The approach quantifies uncertainties in neural network training.

Abstract

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.