Abstract
Although the high-level Keras application programming interface (API) of TensorFlow makes it is easy to build and train neural network models, it still has limitations. For example, not all the new network layers or loss functions proposed by the authors in the latest papers are included in the high-level API, so these layers and loss functions are not available to use for building new network models. In this chapter, we introduce advanced TensorFlow techniques to show how to customize the API for specific purposes, such as custom network layer, custom loss function, custom metric function, and custom callback function. Then, we provide two scenarios of building neural networks for multi-category classification. In the first scenario, we apply the high-level Keras API, and in the second, we use the custom API of TensorFlow.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call