| |
 |
UNIVERSITY OF BUCHAREST
FACULTY OF PHYSICS Guest
2026-06-11 23:58 |
 |
|
|
|
Conference: Bucharest University Faculty of Physics 2026 Meeting
Section: High-Energy Physics
Title:
Stable Training of EPiC-GAN for High-Energy
Physics Jet Generation
via Exponential Moving Average and
Distribution-Aware Checkpointing
Authors:
Dan Adrian MOTOC (1)
Affiliation:
Faculty of Physics, West University of Timisoara
E-mail
dan.motoc94@e-uvt.ro
Keywords:
Abstract:
Generative models offer a promising route to accelerating Monte Carlo simulation in
high-energy physics (HEP). We present a systematic study of generative models for top-quark
jet synthesis on the JetNet-30 benchmark, progressing from a flat MLP WGAN-GP baseline to
a faithful reimplementation of EPiC-GAN [1]. We identify and document a training instability
inherent to point-cloud GANs trained with LSGAN: the generator loss does not correlate
with distribution quality, causing the generator to oscillate between reproducing particle-level
features and jet multiplicity across training checkpoints — a phenomenon we term checkpoint
oscillation. We propose two complementary fixes: (1) Exponential Moving Average (EMA)
of generator weights with decay τ = 0.999, which eliminates oscillation by smoothing the
weight trajectory, and (2) distribution-aware checkpointing, which selects the best model
by the sum of Wasserstein distances over all target observables including multiplicity. Our
final model achieves W η
1 = 9.6 × 10−4, W φ
1 = 1.0 × 10−3, W pT
1 = 3.6 × 10−4, multiplicity
W1 = 0.038, and W M
1 = 2.49 × 10−3 on the JetNet-30 top-quark test set, with particle-level
W P
1 = 0.93 × 10−3 exceeding the published EPiC-GAN (2.1 × 10−3). All results are from
a single training seed without error bars.
References:
[1] E. B¨uhmann, G. Kasieczka, J. Thaler, EPiC-GAN: Equivariant Point Cloud Generation for Particle Jets, SciPost Phys. 15, 130 (2023), arXiv:2301.08128
|
|
|
|