How Deep Learning is Revolutionizing Sports Analytics
summary
Predicts a rusher’s final yardage at handoff from NFL shoulder‑sensor tracking by encoding 21 players into tensors, applying convolutions and pooling, and outperforming baselines; model goes live in NFL dashboards and powers a rushing yards over expected metric. Builds a multi‑stage concussion pipeline: helmet detection → 3D conv impact spotting → Hungarian matching to map video boxes to tracking IDs, then aggregate contacts per player. For soccer, fine‑tunes ImageNet backbones with temporal‑channel stacking (T−1,T,T+1) and 3D convs to detect events; competitions yield top placements and production adoption.
AI-written summary · about
chapters (12)
- 0:00From Data to Performance
- 0:58What is Kaggle?
- 3:13About me
- 6:20Deep Learning in Sports Analytics - Key Impacts
- 12:15Sports analytics competitions
- 13:28NFL Big Data Bowl 2020
- 18:17Final standings
- 22:59NFL 1st and Future-Helmet Assignment
- 26:34DFL-Bundesliga Data Shootout
- 28:18Model design
- 30:55NFL 1st and Future-Player Contact Detection
- 32:21Summary
related talks
33mUsing machine learning to improve player safety in the NFL
35mFootball video analysis using Deep Learning
38mDeep Learning plays Handball
29mHow the NFL builds computer vision training datasets at scale
41mThe evolution of basketball with data science
27mAnalysis of sports scenes using computer vision models
42mDeep Learning for Automatic Basketball Video Production
37mRecent Solutions Using Deep Learning Models/ The How and Why