front matter
A deep learning system can be assumed to be efficient if it can bridge two different worlds—research and prototyping with production operations. Teams who design such systems must be able to communicate with practitioners across these two worlds and work with the different sets of requirements and constraints that come from each. This requires a principled understanding of how the components in deep learning systems are designed and how they are expected to work in tandem. Very little of the existing literature covers this aspect of deep learning engineering. This information gap becomes an issue when junior software engineers are onboarded and expected to become effective deep learning engineers.
Over the years, engineering teams have filled this void by using their acquired experience and ferreting out what they need to know from the literature. Their work has helped traditional software engineers build, design, and extend deep learning systems in a relatively short amount of time. So it was with great excitement that I learned that Chi and Donald, both of whom have led deep learning engineering teams, have taken the very important initiative of consolidating this knowledge and sharing it in the form of a book.
We are long overdue for a comprehensive book on building systems that support bringing deep learning from research and prototyping to production. Designing Deep Learning Systems finally fills this need.