Skip to content
DraconiticResearch
Search Sources API
209,932 studies31 sourcesUpdated 10 min ago
PreprintCC BY 4.0arXiv

Higher Order Component Attribution via Learned Surrogates in High Energy Physics

  • Lauri Laatu
  • Aritra Bal
  • Benedikt Maier
  • Markus Klute
  • Michael Spannowsky
Open at sourcePDFDOICite

Abstract

Experimental High Energy Physics (HEP) has adopted the transformer architecture for identification of jets, particularly the state-of-the-art Particle Transformer (ParT) which injects explicit information about particle correlations into the attention mechanism. While able to achieve state-of-the-art results in accuracy, interpretability to its effectiveness has been limited. In this paper we propose component attribution methods using Kolmogorov-Arnold Networks (KANs) that directly tie back to the physical descriptors of jets, to probe whether mechanisms such as Multi-Head Attention (MHA), or the ParT architecture, are able to learn these as discriminating features.

Related

  1. Polyconvex neural-network surrogates with learned paired directional features for anisotropic hyperelastic homogenization
    S. Ares de Parga, A. Cornejo, J. A. Hernández +1
    arXivPreprintCC BY 4.0Full textNumerical analysis & scientific computingComputational physicsarXiv
  2. Learned Surrogates of Three-Dimensional Magnetohydrodynamic Turbulence
    Semih Kacmaz, E. A. Huerta
    arXivPreprintarXiv licenceFull textInstrumentation & methodsFluid dynamicsarXiv
  3. Structured Extrema Errors in Classical Surrogates for Viscous Burgers: A Physics-Consistent Interpretation
    Youssef Oubari
    arXivPreprintCC BY 4.0Full textFluid dynamicsarXiv
  4. Hybrid modeling with attribution-guided symbolic enhancements of known physics: Applications to complex biological and engineering systems
    Akshay Kudva, Hangjun Cho, Parth Shah +2
    Chemical Engineering SciencePeer-reviewedCC BY6 citedStatistical & nonlinear physicsOpenAlex
  5. Type 1.5 vortex bags in single component Ginzburg-Landau models with higher order potentials
    Martin Speight, Thomas Winyard
    arXivPreprintCC BY 4.0Full textCondensed matterarXiv

Cite

BibTeXRIS
BibTeX
@misc{laatu2026higher,
  title = {{Higher Order Component Attribution via Learned Surrogates in High Energy Physics}},
  author = {Lauri Laatu and Aritra Bal and Benedikt Maier and Markus Klute and Michael Spannowsky},
  howpublished = {arXiv},
  year = {2026},
  month = nov,
  doi = {10.48550/arxiv.2611.08656},
  eprint = {2611.08656},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.08656},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - Higher Order Component Attribution via Learned Surrogates in High Energy Physics
AU  - Lauri Laatu
AU  - Aritra Bal
AU  - Benedikt Maier
AU  - Markus Klute
AU  - Michael Spannowsky
PY  - 2026
DA  - 2026/11/05
PB  - arXiv
DO  - 10.48550/arxiv.2611.08656
UR  - https://arxiv.org/abs/2611.08656
AB  - Experimental High Energy Physics (HEP) has adopted the transformer architecture for identification of jets, particularly the state-of-the-art Particle Transformer (ParT) which injects explicit information about particle correlations into the attention mechanism. While able to achieve state-of-the-art results in accuracy, interpretability to its effectiveness has been limited. In this paper we propose component attribution methods using Kolmogorov-Arnold Networks (KANs) that directly tie back to the physical descriptors of jets, to probe whether mechanisms such as Multi-Head Attention (MHA), or the ParT architecture, are able to learn these as discriminating features.
ER  -