pith. sign in

arxiv: 2507.21214 · v2 · submitted 2025-07-28 · ✦ hep-ph · hep-ex

The Physics Behind ML-based Quark-Gluon Taggers

Sophia Vent , Ramon Winterhalder , Tilman Plehn This is my paper

Pith reviewed 2026-05-19 01:50 UTC · model grok-4.3

classification ✦ hep-ph hep-ex
keywords quark-gluon taggingmachine learning interpretabilityShapley valuessymbolic regressionjet substructureexplainable AIhigh-energy physics
0
0 comments X

The pith

Machine learning quark-gluon taggers can be explained by extracting physics-linked latent features and deriving compact approximation formulas via symbolic regression.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines ways to interpret powerful but opaque machine learning tools for distinguishing quark jets from gluon jets. It first locates the main hidden features inside the ML model that align closely with known physics quantities, using both straightforward linear checks and more complex non-linear ones. It next applies Shapley values to rank which features matter most, while pointing out that the usual calculation assumes inputs are unrelated and can give misleading results when real correlations exist. Finally, symbolic regression turns the ML decisions into short mathematical formulas that reproduce the tagger's behavior. This sequence aims to connect black-box outputs back to understandable physical mechanisms in high-energy collisions.

Core claim

For quark-gluon tagging, leading latent features that correlate strongly with physics observables are identified through linear and non-linear approaches. Shapley values assess feature importance, although standard implementations assume independent inputs and can distort attributions when correlations are present. Symbolic regression then derives compact formulas that approximate the tagger output.

What carries the argument

Symbolic regression applied to ML tagger outputs, after first identifying latent features correlated with physics observables and assessing their importance with Shapley values.

If this is right

  • Physics observables become usable to trace the internal decisions of ML jet taggers.
  • Feature importance rankings become more reliable once correlations among jet properties are accounted for.
  • Compact formulas can replace full ML models in some analyses while keeping most of the discrimination power.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could be tested by checking whether the formulas recover known differences in quark and gluon radiation patterns predicted by QCD.
  • Similar latent-feature plus symbolic-regression steps might clarify other ML applications in collider physics where correlations are common.
  • Accounting for input correlations in Shapley calculations may require physics-specific adaptations rather than off-the-shelf tools.

Load-bearing premise

Symbolic regression applied to the ML tagger outputs will produce compact formulas that retain sufficient accuracy and physical interpretability to be useful approximations.

What would settle it

Apply the derived symbolic formulas to a fresh set of simulated quark and gluon jets and measure how closely their tagging efficiency and purity match the original ML model; substantial drops in performance would show the approximations are inadequate.

read the original 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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript proposes a methodological pipeline for interpreting ML-based quark-gluon jet taggers. It first identifies leading latent features that correlate strongly with physics observables using both linear and non-linear approaches. It then applies Shapley values to assess feature importance, while noting that the standard implementation assumes independent inputs and may produce distorted attributions in the presence of correlations. Finally, it employs symbolic regression to derive compact formulas that approximate the tagger output.

Significance. If the outlined methods are carried through with concrete validation, the work has the potential to improve physical interpretability of black-box ML models in a challenging high-energy physics application. By combining feature correlation analysis, correlation-aware attribution, and symbolic regression, it could yield both diagnostic insights into quark-gluon discrimination and practical, human-readable approximations to existing taggers.

major comments (1)
  1. [Abstract] The abstract describes the intended methods but supplies no quantitative results, validation metrics, error analysis, or implementation details. Without these, it is impossible to determine whether the linear/non-linear latent-feature identification, the Shapley-value attributions, or the symbolic-regression approximations actually support the stated goals of physical insight and useful accuracy for quark-gluon tagging.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and for the constructive feedback. We appreciate the positive assessment of the potential significance of the work and address the major comment below.

read point-by-point responses

  1. Referee: [Abstract] The abstract describes the intended methods but supplies no quantitative results, validation metrics, error analysis, or implementation details. Without these, it is impossible to determine whether the linear/non-linear latent-feature identification, the Shapley-value attributions, or the symbolic-regression approximations actually support the stated goals of physical insight and useful accuracy for quark-gluon tagging.

    Authors: We agree that the abstract would benefit from including key quantitative highlights to allow readers to immediately gauge the strength of the results. The full manuscript contains detailed validation of the latent-feature correlations (both linear and nonlinear), correlation-aware Shapley attributions, and the fidelity of the symbolic-regression approximations. In the revised version we will add a concise sentence or two to the abstract summarizing the main numerical outcomes, such as the leading correlation values and the approximation accuracies, while preserving the abstract's brevity and focus. revision: yes

Circularity Check

0 steps flagged

No significant circularity in methodological description

full rationale

The abstract outlines a standard pipeline of applying established ML explainability tools—latent feature identification (linear and non-linear), Shapley-value attribution with an explicit caveat on input correlations, and symbolic regression to approximate tagger outputs—without presenting any equations, derivations, fitted parameters, self-citations, or uniqueness claims. No step reduces by construction to its own inputs or renames a result as a prediction. The work is a direct methodological application to quark-gluon tagging and remains self-contained as a description of techniques rather than a closed derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; assessment is limited to the high-level description provided.

pith-pipeline@v0.9.0 · 5576 in / 1273 out tokens · 51561 ms · 2026-05-19T01:50:01.440062+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Dissecting Jet-Tagger Through Mechanistic Interpretability

    hep-ph 2026-05 accept novelty 8.0

    A Particle Transformer jet tagger contains a sparse six-head circuit whose source-relay-readout structure recovers most performance and whose residual stream preferentially encodes 2-prong energy correlators.

  2. Explainable AI for Jet Tagging: A Comparative Study of GNNExplainer, GNNShap, and GradCAM for Jet Tagging in the Lund Jet Plane

    hep-ph 2026-04 unverdicted novelty 5.0

    Explainability techniques applied to LundNet show that assigned node importance correlates with classical jet substructure observables such as N-subjettiness ratios and energy correlation functions, with shifts across...