Hands-On Machine Learning with TensorFlow.js

Hardware acceleration using web browser APIs

One possible drawback of using a web browser as a machine learning platform is performance. Web browsers are usually single-process applications. Typically, they don't work with CPU-intensive use cases such as machine learning. However, there are some ways we can achieve high performance in web browsers. Modern web browsers provide standard APIs that use local hardware accelerators, such as GPUs. For example, WebGL is a standardized specification that allows us to use GPUs from web browsers with ease. By making use of such APIs, TensorFlow.js can achieve competitive performance, even in web browsers.

Although WebGL was originally developed for graphical processing, TensorFlow.js has its own wrapper ...