Explore the evolution of machine learning systems at Pinterest in this 46-minute conference talk. Gain insights into the company's growth, its use of ML in the home feed ranking model, and the mobile development lifecycle. Learn about crucial decisions in collecting training data, automated training processes, and the implementation of A/B experiments. Discover how Docker Compose and a debug framework enhance development velocity. Conclude with a demonstration and Q&A session to deepen your understanding of production ML systems in a large-scale social media platform.
Accelerating Development Velocity of Production ML Systems with Docker
Overview
Syllabus
Introduction
What is Pinterest
Evolution of Pinterest
Machine Learning at Pinterest
How many people use Pinterest
Pinterest Home Feed
Ranking Model
Ranking Model Evolution
Current Model
Mobile Development Lifecycle
Collecting Training Data
Important Decisions
Labels and Sampling
Automated Training
ROC AUC
A B Experiment
Food analogy
Dependency Health
Debug Framework
Docker Compose
Edward Murphy
Demo
Where to find me
QA Time
Taught by
Docker
