ABSTRACT
This chapter explores “machine learning” as a synonym for artificial intelligence, because of the dominance of this form of Artificial Intelligence (AI) workload. Many traditional workloads have extremely undemanding latency requirements, such as batch processing of server logs, but AI makes it easier to build applications that interact with the physical world, where delayed results can have consequences that are a little more serious than a spinning hourglass. Latency requirements may be less clear-cut in, say, a retail setting, where a model might be interacting with a customer or trying to target offers to specific demographics. The solution has been to create content delivery networks, placing bandwidth-heavy content as close to every user as possible through purpose-built infrastructure, managed by specialized providers like CloudFlare. Google has also created a second open-source project called KubeFlow, which uses both TensorFlow and TensorFlow Serving as components.
