Distributed Deep Learning with Docker at Salesforce

Explore the challenges and solutions of deploying deep learning models in a production environment at Salesforce. Learn how Docker enables scalable system design, simplifies development, and facilitates efficient testing. Discover the benefits of using Docker for distributed deep learning, including its role in focusing on service interactions rather than hardware complexities. Gain insights into designing systems for team specialization, addressing throughput concerns, and serving deep learning models effectively. Understand how Docker simplifies the lives of developers and data scientists by allowing them to build and test end-to-end systems on local machines. This 33-minute talk by Jeff Hajewski from Salesforce covers topics such as deep learning review, designing for team specialization, throughput considerations, model serving, and the advantages of Docker in simplifying complex systems.

Intro
Caveats
Deep Learning Review
Deep Learning at Salesforce
Designing for team specialization
What about throughput?
Serving deep learning models
Interacting with the model server
This solves additional challenges
Testing
Docker simplifies our lives