The Physics Behind ML-based Quark-Gluon Taggers

TL;DR

This paper explores the physics principles behind ML-based quark-gluon taggers, using explainability methods like feature importance and symbolic regression to understand and approximate their behavior.

Contribution

It introduces a physics-informed analysis of ML taggers, applying Shapley values and symbolic regression to interpret and simplify their decision-making process.

Findings

Identified key latent features correlating with physics observables.

Demonstrated limitations of standard Shapley values due to input correlations.

Derived compact formulas approximating the ML tagger outputs.

Abstract

Jet taggers provide an ideal testbed for applying explainability techniques to powerful ML tools. For theoretically and experimentally challenging quark-gluon tagging, we first identify the leading latent features that correlate strongly with physics observables, both in a linear and a non-linear approach. Next, we show how Shapley values can assess feature importance, although the standard implementation assumes independent inputs and can lead to distorted attributions in the presence of correlations. Finally, we use symbolic regression to derive compact formulas to approximate the tagger output.