Quantum Attention for Vision Transformers in High Energy Physics

  • 2024-11-20 18:11:17
  • Alessandro Tesi, Gopal Ramesh Dahale, Sergei Gleyzer, Kyoungchul Kong, Tom Magorsch, Konstantin T. Matchev, Katia Matcheva
  • 0

Abstract

We present a novel hybrid quantum-classical vision transformer architectureincorporating quantum orthogonal neural networks (QONNs) to enhance performanceand computational efficiency in high-energy physics applications. Building onadvancements in quantum vision transformers, our approach addresses limitationsof prior models by leveraging the inherent advantages of QONNs, includingstability and efficient parameterization in high-dimensional spaces. Weevaluate the proposed architecture using multi-detector jet images from CMSOpen Data, focusing on the task of distinguishing quark-initiated fromgluon-initiated jets. The results indicate that embedding quantum orthogonaltransformations within the attention mechanism can provide robust performancewhile offering promising scalability for machine learning challenges associatedwith the upcoming High Luminosity Large Hadron Collider. This work highlightsthe potential of quantum-enhanced models to address the computational demandsof next-generation particle physics experiments.

 

Quick Read (beta)

loading the full paper ...