Opening Keynote
summary
Overview of concrete claims, tools and results: AWS reports shipping Few‑Shot Learning into Rekognition and Lookout for Vision so customers train with 10 images per label or ~30 images total for defect detection; a mock factory proves synthetic anomalies work for Dafgards. A new OCR model, SCATTER improves irregular text accuracy by 3.7% and appears in Textract/Recognition. SageMaker runs hundreds of billions of predictions/month, adds multi‑model endpoints hosting up to 1000 models per instance, Clarify uses Shapley values (10x faster) for explainability, and Inferentia cuts inference cost up to 70% while raising throughput.
AI-written summary · about
chapters (11)
- 0:00Welcome and event overview
- 1:55Machine learning's explosive growth
- 3:45Customer-obsessed science explained
- 5:51Few-shot learning for less data
- 8:57Irregular text recognition with SCATTER
- 15:02Computational humor with Alexa
- 30:15ML services from Amazon to AWS
- 35:56Scaling ML with Amazon SageMaker
- 48:29Training developers in ML
- 50:18Intuit's AI-driven platform
- 59:37Closing remarks
related talks
29mAmazon Lookout for Vision
30mMachine learning for computer vision in pharmaceutical manufacturing
01h 05mKeynote: Builders Day
27mComputer vision for automated quality inspection
40mMachines are Learning
20mIndustrial-Scale Machine Learning
51mProfundizando en Aprendizaje Automático para Desarrolladores
01h 30mDriving growth with innovative digital products