Key takeaways
- What's new in H2O AI Cloud (now the H2O Sovereign AI Platform), including single sign-on API tokens, H2O Drive for centralized data import, and a redesigned homepage
- How H2O Feature Store works, from registering feature sets to online and offline stores, demonstrated by ingesting weather and flight data to predict delays
- Driverless AI's recipe management and Shapley reason codes, which let users see and edit the exact python code and features behind a model
- New H2O-3 algorithms including uplift modeling for treatment effects, RuleFit for interpretable rules, and Infogram for feature selection
- Hydrogen Torch's no-code deep learning additions: instance segmentation, Grad-CAM explainability, and audio classification via spectrograms
- MLOps updates like MLflow model registry support and admin analytics, plus H2O Wave's new snippets, themes, and server-side tables for large data sets
Overview
This recording of H2O.ai's first Product Day walks through what shipped that spring across H2O AI Cloud (now the H2O Sovereign AI Platform), from platform updates like single sign-on API tokens and H2O Drive to the newly launched Feature Store and Document AI. The team demos Driverless AI's recipe management, new H2O-3 algorithms like uplift modeling and RuleFit, and the no-code Hydrogen Torch deep learning product. It also covers MLOps additions like MLflow registry integration, plus what's new in H2O Wave and the App Store. The session closes with two healthcare apps: a COVID-19 hospital occupancy simulator and a gene mutation risk model.