Uber has been at the leading edge of productizing and democratizing machine learning since 2015. Every aspect of the Uber experience is powered by data and ML - everything from in-app ETAs, menu recommendations, and map labeling to driver dispatch and customer support. In this talk, we want to discuss the ML Platforms, use-cases, and the data infrastructure we plan to build for 2019 and beyond.
ML platforms at Uber enables internal teams to seamlessly build, deploy, and operate machine learning solutions at Uber’s scale. It is designed to cover the end-to-end ML workflow: manage data, train, evaluate, and deploy models, make predictions, and monitor predictions.
- Python Model Serving at Scale (PyML)
- Hyperparameter Optimization (AutoTune)
- Deep Learning in the Cloud (GCP)
- Visual Drag-and-drag Data Preparation and Feature Generation pipelines (uWorc)
- Knowledge Feed of Notebooks, Dashboards, Experimentation Results
- Real-time pipelines (powered by Apache Flink and Cassandra)
- Challenges of reliability and monitoring at Uber's scale
- Concrete use cases in production and new ones coming down the pike
(Will be co-presenting with Mike Del Balso)