Overview
iGAPT is a novel transformer-based generative model for high-energy physics simulations, offering improved efficiency and accuracy in particle collision modeling through induced attention mechanisms.
Highlights
- Developed an induced particle-attention mechanism for efficient particle cloud generation
- Achieved linear time complexity (O(n)) compared to quadratic (O(n²)) in MPGAN
- Integrated global jet attribute conditioning for improved physics fidelity
- Surpassed MPGAN performance across multiple evaluation metrics
Architecture
- Induced Attention: Novel attention mechanism for efficient particle interactions
- Global Conditioning: Integration of physics-based global attributes
- Transformer Blocks: Modified transformer architecture for particle data
- GAN Framework: Adversarial training setup for realistic generation