Physics at the LHC: Data Analysis and Global Network Challenges with Exascale Data
summary
Large LHC datasets arrive at 40 MHz crossings with up to ~30 interactions now and 140–200 planned for the HL‑LHC, read out by 100 million channels every 25 ns and reduced by hardware plus software triggers to ~1 kHz output, producing tens of petabytes per year. A worldwide Tier0/Tier1/Tier2/Tier3 grid and 150–300 Gbps science networks scale poorly (historic growth ~72%/yr, recent >100%), so the program adopts ML: boosted decision trees yield a 1.8× sensitivity gain in H→γγ (contributing to 4σ→6σ), CNNs improve jet/W tagging, and generative models and neuromorphic pilots target simulation and tracking speedups.
AI-written summary · about
chapters (10)
- 0:00LHC physics and data challenges
- 6:43LHC data volume and collision complexity
- 12:46Global data flow and computing model
- 21:18Higgs discovery and ML in physics
- 30:13ML for data reconstruction
- 35:34Simulation and generative models
- 43:46Neural networks for tracking and tagging
- 50:25Neuromorphic computing and future tech
- 53:13Intelligent global data management
- 56:53Question and answer session
related talks
33mData science in high energy physics
29mStreamlining Competitive Data Science at CERN: Running ML...
35mBuilding a 5-Exaflop Supercomputer for Meta-AI Research and...
01h 17mHPC application in analysis of large scale complex networks
19mA world of data
14mFrom Reservation to 1,600 GPUs: A Petabyte-Scale Ray Data Pipeline
26mLarge Scale Graph Processing & Machine Learning Algorithms
27mTech Talk