Designing Deep Learning Systems

Appendix B. Survey of existing solutions

Deep learning systems are huge undertakings when implemented from scratch. In some situations, special requirements may warrant spending the extra effort to build a deep learning system from scratch. In other cases, given finite resources and time, it may make sense to use existing components, or even systems as a whole, and tailor them to your own needs.

The purpose of this appendix is to examine a few deep learning systems that have been implemented by different cloud vendors and open source communities. These operations range from serverless deployment to custom service container deployment. You will gain a sense of which pieces of these operations you can use to design your own project by comparing ...