Travelport: Near Real-time AI Deployment with Huge Data and Super Low Latency
"Benefit of H2O is that it's really, really fast!"
Principal Data Scientist
Scaling ML Platform
Overview of the Challenge
This talk discusses typical machine learning challenges and solutions for (1) model decay, (2) volume of training data, (3) time series data, (4) compute necessary, (5) scoring latency, and (6) DevOps. Previously, models were accurate at first, but started decaying after only 10 – 15 hours. With near real-time retraining of ML models and strong DevOps processes Travelport solved these challenges.