Secondary vertex reconstruction with MaskFormers

Area of Science:

• High-energy particle physics
• Particle detection and reconstruction
• Machine learning applications in physics

Background:

• Reconstruction of secondary vertices from heavy-flavour hadron decays is vital for jet identification in particle physics.
• Existing methods face limitations in optimization, phase space consistency, and multi-vertex reconstruction.

Purpose of the Study:

• To introduce a novel secondary vertex reconstruction method using object detection and computer vision.
• To overcome limitations of traditional and current machine learning approaches for vertex reconstruction.

Main Methods:

• Developed a single model that directly predicts the presence and properties of multiple secondary vertices.
• Applied the method to simulated proton-proton collision events.

Main Results:

• Achieved a 10% improvement in vertex finding efficiency compared to state-of-the-art methods.
• Enabled accurate vertex fitting for key properties like transverse momentum and flight distance.
• Improved light-jet rejection by 50% and c-jet rejection by 15% at 70% b-jet efficiency.

Conclusions:

• The proposed object detection approach offers significant advancements in secondary vertex reconstruction.
• Demonstrates the potential of computer vision techniques for enhancing high-energy physics experiments.
• Paves the way for more powerful and flexible reconstruction tools.