DeepTreeGANv2: Iterative Pooling of Point Clouds

TL;DR

DeepTreeGANv2 tackles the need for fast, high-fidelity generation of large point clouds representing particle showers in calorimeters by introducing a tree-structured critic that iteratively downscales point clouds via a bipartite pooling mechanism. The critic comprises three subcritics and employs an embedding layer and a central node update to preserve hierarchical information, conditioned on jet kinematics $p_T$, $oldsymbol{ heta}$, and mass. On the JetNet 150 dataset, it achieves competitive metrics relative to state-of-the-art GANs, capturing constituent distributions and jet mass with good fidelity, though the mass peak is not perfectly sharp. The approach scales to 150 constituents and offers a reusable pooling strategy that could benefit other point-cloud tasks in high-energy physics and beyond.

Abstract

In High Energy Physics, detailed and time-consuming simulations are used for particle interactions with detectors. To bypass these simulations with a generative model, the generation of large point clouds in a short time is required, while the complex dependencies between the particles must be correctly modelled. Particle showers are inherently tree-based processes, as each particle is produced by the decay or detector interaction of a particle of the previous generation. In this work, we present a significant extension to DeepTreeGAN, featuring a critic, that is able to aggregate such point clouds iteratively in a tree-based manner. We show that this model can reproduce complex distributions, and we evaluate its performance on the public JetNet 150 dataset.

Figures (2)

• Figure 1: The critic, as described in \ref{['sec:critic']}
• Figure 2: The distributions of $\eta^{\mathrm{rel}}\xspace, \phi^{\mathrm{rel}}\xspace, {p}_{\mathrm{T}}^{\mathrm{rel}}\xspace$ for constituents of the jets and the mass of the jets for the top-quarks JetNet 150 dataset (Simulation), as described in Section \ref{['sec:jetnet']} and DeepTreeGAN (Model).