Dive into a comprehensive 58-minute conference talk by Xavier Amatriain on practical deep learning systems, presented at Full Stack Deep Learning in November 2019. Explore Amatriain's background and insights from his work at Qi, Netflix, and other tech companies. Learn about crucial aspects of deep learning, including data handling, transfer learning, fine-tuning, and the importance of simple models. Discover real-life examples, architecture engineering, and the differences between supervised and self-supervised learning. Examine topics such as data bias, fairness, and deploying models in production. Gain valuable knowledge on evaluation approaches, metrics, and machine learning infrastructure. Conclude with a comparison of deep learning and linear models, followed by a Q&A session.
Xavier Amatriain on Practical Deep Learning Systems - November 2019
Overview
Syllabus
Introduction
Xaviers background
What is Qi
Publications
Lessons Learned
Question
Netflix Price
Meta Metadata
Unreasonable Effectiveness
Netflix example
Data
Transfer Learning
Fine Tuning
Simple Models
Occams Razor
More connections to deep learning
Recommended papers
Real life example
Complex models
Avoid this trap
Feature engineering
Reusable features
Examples
Architecture Engineering
Supervised vs Supervised
Models in Deep Learning
Self Supervision
Insample
Netflix Prize
Deep Models
Data Bias
Bias
Fairness
Models in Production
Models in Other Domains
Evaluation Approach
Metrics
Systems frameworks
Systems and frameworks
Machine learning infrastructure
Machine learning beyond deep learning
Deep learning vs linear models
Conclusions
Questions
Taught by
The Full Stack
