Combining algorithms and quality data to build better AI platforms

on YouTube ↗13m

" summary

Argues that better data and tighter AI–analytics integration beat marginal algorithm gains, and shows a working stack: BigDL running on Spark and Hadoop with 200+ neural layers and feature parity with Caffe/TensorFlow/Torch to unify training, inference and analytics and cut TCO, and Intel Optane DC persistent memory extending system RAM by >3 TB per CPU and delivering concrete wins — Spark SQL ~5x, Cassandra ~9x throughput vs PCIe SSD, SAP HANA restart 50 → 4 minutes.

AI-written summary · about

" chapters (5)

1. 0:00Introduction
2. 0:11Data and AI
3. 2:22Platform AI
4. 5:47Intel Persistent Memory
5. 9:33Diversity and Inclusion

" related talks

• 09mThe AI stack from chip to analyticsMa Ziya · Strata Data Conference 2017 - New York, New York
• 06mUnleashing intelligence and data analytics at scaleMa Ziya · Strata Data Conference 2017 - New York, New York
• 58mAI-Enhanced Big Data: Integrating Private LLMs and Vector DatabasesRohit Bhardwaj · Great International Developer Summit 2025
• 06mCollaboration and openness drive innovation in artificial intelligenceMartin Hall · Strata + Hadoop World 2016 - New York, New York
• 10mThe game-changing future for Data and AIMadhu Kochar · Strata Data Conference 2019 - New York, NY
• 24mTools & Techniques for a cost optimized Analytics & AI pipelineAnand Kulkarni · MachineCon 2022
• 35mHigh-performance data foundation for AI successJason Hunter · AWS Summit Sydney 2025
• 26mThe Hidden Foundation: How Effective Data Management Determines AI System SuccessKeith McClellan · Cloud Native + Kubernetes AI Day North America 2025