Building smarter ML and AI models and making them more accurate
summary
Claims a SQL-based GPU platform called scream that reads from storage straight into GPU cores to avoid memory bottlenecks, runs terabytes–petabytes without copies, and parallelizes queries across tens of thousands of GPU cores. Demonstrated reductions: 3.5 weeks → 1h20m for a Barclays fraud join, 4,000 servers → 16 servers consolidation at SKT, ingestion ~3 TB/hr per GPU, and a Dutch bank case that cuts aggregation latency from 12 hours to 40 minutes and lowers fraud losses by 48%.
AI-written summary · about
chapters (4)
- 0:00Start
- 8:04Traditional Modelling Steps
- 9:30Challenges of the new ERA of data-intensive computing
- 29:52Enterprise Fraud Management (EFM) in Financial Institutions Using ML
related talks
18mThe "New Normal" or An Evolution
27mEnterprise AI at Scale
26mEfficiently serving LLMs at scale.
09mDemocratizing Access to GPUs for Faster MLAI Innovation with Generative AI
27mEnterprise AI at Scale: Success Stories and Insights
24mManipulating Data in an Artificial Intelligence World
46mShorten the Path to Production with In-Database Machine Learning
39mUnlock the potential of your data: Optimize your data store architecture...