We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

hep-ph

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

High Energy Physics - Phenomenology

Title: Does Lorentz-symmetric design boost network performance in jet physics?

Abstract: In the deep learning era, improving the neural network performance in jet physics is a rewarding task as it directly contributes to more accurate physics measurements at the LHC. Recent research has proposed various network designs in consideration of the full Lorentz symmetry, but its benefit is still not systematically asserted, given that there remain many successful networks without taking it into account. We conduct a detailed study on the Lorentz-symmetric design. We propose two generalized approaches for modifying a network - these methods are experimented on PFN, ParticleNet, and LorentzNet, and exhibit a general performance gain. We also reveal for the first time that the underlying reason behind the known improvement the "pairwise mass" feature brings to the network is that it introduces a structure that adheres to full Lorentz symmetry. We confirm that Lorentz-symmetry preservation serves as a strong inductive bias of jet physics, hence calling for attention to such general recipes in future network designs.
Comments: 14 pages, 6 figures
Subjects: High Energy Physics - Phenomenology (hep-ph); High Energy Physics - Experiment (hep-ex); Data Analysis, Statistics and Probability (physics.data-an)
Cite as: arXiv:2208.07814 [hep-ph]
  (or arXiv:2208.07814v1 [hep-ph] for this version)

Submission history

From: Congqiao Li [view email]
[v1] Tue, 16 Aug 2022 15:53:53 GMT (661kb,D)

Link back to: arXiv, form interface, contact.