Related Experiment Video
Updated: Jun 6, 2025
12:08
From Voxels to Knowledge: A Practical Guide to the Segmentation of Complex Electron Microscopy 3D-Data
Published on: August 13, 2014
24.5K
Foundations of automatic feature extraction at LHC-point clouds and graphs
Akanksha Bhardwaj1, Partha Konar2, Vishal Ngairangbam3
1Department of Physics, Oklahoma State University, Stillwater, OH 74078 USA.
The European Physical Journal. Special Topics
|November 28, 2024
Summary
Deep learning is crucial for the Large Hadron Collider (LHC), enhancing simulations and analyses. Physics-inspired deep learning offers advantages beyond expert intuition, improving feature extraction and understanding.
Area of Science:
- High Energy Physics
- Computational Physics
- Machine Learning
Background:
- Deep learning algorithms are increasingly vital for the Large Hadron Collider (LHC) experiments.
- These algorithms enhance detector simulations and physics analyses, including searches for physics beyond the Standard Model.
- The ability of deep learning to handle high-dimensional data is transformative, potentially automating the design of physics-intuitive variables.
Purpose of the Study:
- To systematically review automatic feature extraction in particle physics from a phenomenological perspective.
- To explore the motivations behind developing physics-inspired deep learning architectures for collider physics.
- To discuss the benefits of incorporating prior physics knowledge into feature extraction, particularly using point cloud representations and graph-based methods.
Main Methods:
- Systematic review of existing research on automatic feature extraction in the context of LHC phenomenology.
- Analysis of the advantages of physics-inspired feature extractors compared to purely data-driven approaches.
- Exploration of point cloud representations and graph-based neural networks for LHC data analysis.
Main Results:
- Physics-inspired feature extraction offers significant advantages, including improved interpretability and qualitative understanding of extracted features.
- Prior knowledge from physics naturally leads to effective representations, such as point clouds, for collider data.
- Graph-based methods show promise for advanced applications in LHC phenomenology.
Conclusions:
- Deep learning, particularly physics-inspired architectures, is essential for advancing LHC research.
- Integrating domain knowledge enhances the effectiveness and interpretability of machine learning models in high energy physics.
- Future work should focus on leveraging graph-based approaches and point cloud representations for sophisticated LHC data analysis.
