What's New in TensorFlow 2.0

Advanced model building using the functional API

As machine learning tasks grow in sophistication, models with multi-stage inputs and outputs become increasingly common. A sizable chunk of real-world use cases involve models with multi-stage inputs and outputs. An example of a real-world model with multiple inputs is a text classification model that looks at both words and character sequences in the input text.

While the Sequential API does a very good job of combining layers in a serial fashion, it cannot be used for describing parallel compositions of underlying layers. In general, it cannot be used to build layer graphs that do not have a linear topology. Its utility is also restricted in cases when a particular layer needs to be utilized ...