Abstract: Keras, the popular deep learning library, is now becoming multi-backend with support for TensorFlow, JAX, and PyTorch. Available now, Keras Core allows developers to create models with all of the simple high-level components of Keras while interchanging between frameworks to take advantage of the benefits of each.
Existing users of each framework can seamlessly integrate their code, including backend-specific code, with Keras. Additionally, Keras Core’s ops suite, which includes the NumPy API, allows you to develop custom components for use on any framework. For current users of tf.keras, Keras Core with the TensorFlow backend serves as a drop-in replacement, meaning that changing your imports is all that is necessary to start taking advantage of the framework-agnostic future.
A unifying library for machine learning, Keras Core enables developers to maximize performance, broaden data source availability, and reach the entire OSS market through distribution across frameworks.
Attendees will learn how to use multi-backend Keras (Keras Core) with several existing machine learning frameworks (TensorFlow, JAX, and PyTorch) and the library's distinct advantages.
Bio: Neel is currently an engineer on the Keras team at Google. His work has been focused on open source development, adding major features to the latest releases of TensorFlow and Keras. Experienced broadly from data science to ML infrastructure, he is responsible for the saving and export of ML models, as well as a developer for cross-framework compatibility at Google. As a key maintainer of keras.io and tensorflow.org, Neel is passionate about educating the data science community at large and helping the latest innovations of AI reach everyone.