Keynote – Joseph Sirosh
summary
Shows concrete pipelines and numbers: uses MMLSpark and transfer learning on ResNet to raise snow‑leopard detection from ~68% to ~89% accuracy, runs featurization distributed on Spark and packages inference in Docker, demonstrates 24 FPGAs serving ~15,700 documents/sec vs ~80/sec on CPUs, embeds R/Python and CNTK in SQL Server 2017 to featurize ~250k images in under an hour (30× faster with GPUs) and exposes production APIs via Azure Cognitive Services.
AI-written summary · about
chapters (13)
- 0:00Fraud detection to AI leadership
- 1:48Data science grunt work
- 4:22ImageNet sparked deep learning
- 8:53Intelligence in video
- 11:25Snow leopard detection demo
- 17:08Hardware for deep neural networks
- 21:27AI development pipeline recap
- 23:37AI with development tools
- 28:11AI inside the database
- 33:05Medical imaging AI demo
- 39:57Democratizing AI with APIs
- 43:19Building AI apps in minutes
- 52:20Three revolutions shaping future
related talks
51mSpotlight on Emerging Data Science Worskspaces and Spark as Development Levers
44mProcessing large data with Apache Spark
37mKeynote – Carol Willing
24mSmart Cities Unleashed
01h 09mNotebook to Pipeline: Hands-On Data Engineering w Python
48mKeynote – Sameer Singh
26mOne Pipeline to Rule Them All: Unifying Multimodal and AI Data Processing with Daft
01h 00mData, visualization, and designing AI