hep-phNov 19, 2026

Who Moved My ZZ Boson? Revealing Resonance Features Inside a Neural Network

Authors: Kevin Alberto Urquía Calderón, Midori Kato, Inar Timiryasov, Oleg Ruchayskiy

Organizations: Niels Bohr Institute, University of Copenhagen, Jagtvej 155A, DK-2200, Copenhagen, Denmark.

Abstract

Methods used in mechanistic interpretability have revealed human-aligned concepts inside language and image models. In this paper, we search for physics concepts inside a neural a generative collider physics model trained on real LHC data (ShellFlow). We discover features inside ShellFlow that represent resonances, such as the ZZ, J/ψJ/ψ, and ΥΥ. The network was never told which muons come from a resonance, it was only trained to generate the kinematics of the final-state particles. We use these features to alter aspects of the ZZ boson: we change its mass between ∼80\sim80 and 105105 GeV, and removing such a feature from the neural network produces a broader ZZ peak. In both cases the rest of the invariant mass spectrum remains mostly unchanged. The feature was built from the gradients of the neural network, and its selectivity was built using pattern recognition methods. A generative particle physics model trained on unlabeled data can be modified internally to change a physical quantity.

Figures & tables

Appendix figures & tables21 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. Dissecting Jet-Tagger Through Mechanistic Interpretability

    May 11, 2026Saurabh Rai, Sanmay GangulyTransformer InterpretabilityAttention Head Analysis

  2. Learning Standard Model structure from LHC data with Riemannian flow matching

    Jul 17, 2026Midori Kato, Kevin A. Urquía-Calderón, Inar Timiryasov +1Constrained Generative ModelingFlow Matching

  3. Uncovering Hidden Systematics in Neural Network Models for High Energy Physics

    May 8, 2026Lucie Flek, Philipp Alexander Jungs, Akbar Karimi +6Uncertainty QuantificationAdversarial Robustness