Data Science at Uber - Full Stack Deep Learning - August 2018

Explore the lifecycle of deep learning models at Uber in this 44-minute conference talk by Jai Ranganathan, former VP of AI and Data at Uber. Gain insights into customer-centric approaches, ticket complexity analysis, and model exploration processes. Learn about data modeling types, cost-benefit tradeoffs, and final architecture considerations. Discover techniques for model validation, open-source visualization, and entity recognition. Delve into challenges such as distributed training with Spark, testing strategies, and performance metrics. Understand the importance of monitoring, retraining pipelines, and managing training data in a large-scale AI environment.

Intro
Jais background
Lifecycle of a deep learning model
Customer obsession ticket resistant
Ticket complexity
Too many transitions
First step exploration
The process
First things first
The problem
Summary
Emily Model
Data
Model Types
Cost Benefit Tradeoffs
Final Architecture
Model Validation
Entity in Wedding
Open Source Visualization
End Result
Challenges
Spark
Distributed training
Testing strategy
Metrics
Department Summary
Monitoring Training
Retraining
Pipeline
Training Data
Summary of Monitoring