Ultra-high-granularity detector simulation with intra-event aware generative adversarial network and self-supervised...

Area of Science:

• Particle Physics
• Artificial Intelligence
• High-Energy Physics

Background:

• Simulating high-resolution detector responses is computationally demanding.
• Existing generative models struggle with the correlated, fine-grained data in ultra-high-granularity detectors.

Purpose of the Study:

• To develop an efficient method for simulating ultra-high-granularity detector responses.
• To address limitations in current generative models for complex detector data.

Main Methods:

• Introduced the Intra-Event Aware Generative Adversarial Network (IEA-GAN).
• Utilized a Transformer-based Relational Reasoning Module for event approximation.
• Implemented Self-Supervised intra-event aware and Uniformity loss functions.

Main Results:

• IEA-GAN generates contextualized, high-resolution detector responses.
• Achieved enhanced sample fidelity and diversity in simulations.
• Successfully applied to the Belle II Pixel Vertex Detector (PXD) with millions of channels.

Conclusions:

• IEA-GAN offers a novel approach to high-granularity detector simulation.
• This method has potential applications in Foundation Models for future colliders like HL-LHC.
• Enables advancements in simulation-based inference and density estimation.