Graph Network in High Energy Physics
summary
Graph networks convert detector hits into nodes and predict edge-by-edge tracking via binary edge scores to build charged-particle trajectories, reducing the costly combinatorics of traditional tracking. A particle-level graph with GRU-style message passing classifies nodes to reject pileup and shows improved jet energy resolution versus prior algorithms; current tests target ~40 simultaneous interactions with projections up to 200 pileup collisions. On jet ID a graph-level classifier outperforms image and sequence baselines on standard benchmarks.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:13Outline
- 0:58Colliding Hadrons
- 1:23Size Of The Challenge
- 2:05CMS Detector
- 2:33What is an Event
- 3:28A Journey Through Matter
- 4:00@ From RAW to High Level data
- 5:16Image and Sequences
- 7:43Tracker Hit Graph
- 8:27Node & Edge Representations
- 10:25Graph Construction
- 10:54Updating Rule
- 11:19Pile-up Classifier
- 11:31Model Performance
- 12:50Jet-id with Interaction Network
- 13:45Classification Performance
- 14:24Summary and Outlooks
related talks
01h 00mPhysics at the Large Hadron Collider: A New Window on Matter, Spacetime and the Universe II
45mApplications of Graph in Fraud Detection
23mUsing Graph Nets (GNs) to predict molecular properties
16mPreview: Graph NNs
39mUsing Graph Neural Networks to Embrace the Dependency Within Your Data
01h 01mToward Universal Graph Representations: Foundations and Frontiers of Graph Foundation Models
31mGraph-Based Machine Learning with Python
01h 23mGraph Neural Networks in Drug Discovery