Keras is an open-source library that provides a Python interface for artificial neural networks. Designed to enable fast experimentation with deep neural networks, it focuses on being user-friendly, modular, and extensible. The library was first released as independent software in 2015, later integrated into the TensorFlow library, and eventually expanded to support additional backends. Keras 3, a full rewrite, serves as a low-level cross-framework language for developing custom components such as layers, models, or metrics that can be used in native workflows in JAX, TensorFlow, or PyTorch with one codebase. As of TensorFlow 2.16, Keras 3 is the default version, though Keras 2 remains available for legacy use.
History
The name 'Keras' derives from the Ancient Greek word κέρας (Keras), meaning 'horn'. The library was developed as part of the research effort of project ONEIROS (Open-ended Neuro-Electronic Intelligent Robot Operating System). Its primary author and maintainer is François Chollet, who was a Google engineer until leaving the company in 2024. Chollet is also the author of the Xception deep neural network model. Up until version 2.3, Keras supported multiple backends, including TensorFlow, Microsoft Cognitive Toolkit, Theano, and PlaidML. From version 2.4 up until version 3.0, only TensorFlow was supported. Starting with version 3.0, as well as its preview version Keras Core, the library became multi-backend again, supporting TensorFlow, JAX, and PyTorch, with additional support for OpenVINO.
Features
Keras contains numerous implementations of commonly used neural-network building blocks, such as layers, objectives, activation functions, optimizers, and tools for working with image and text data. These simplify programming for deep neural networks. The code is hosted on GitHub, and community support forums include the GitHub issues page. In addition to standard neural networks, Keras supports convolutional and recurrent neural networks. It includes common utility layers like dropout, batch normalization, and pooling. The library allows users to produce deep models on smartphones (iOS and Android), on the web, or on the Java Virtual Machine. It also enables distributed training of deep-learning models on clusters of graphics processing units (GPU) and tensor processing units (TPU).
Adoption and Impact
Keras became a widely adopted high-level API within the Deep learning community, particularly after its integration into TensorFlow as the primary interface for model building. Its emphasis on simplicity and rapid prototyping made it a common entry point for practitioners in Machine learning and Artificial intelligence. The library's design influenced subsequent frameworks and contributed to the broader accessibility of Neural network development. By supporting multiple backends in version 3.0, Keras aimed to provide a unified interface across different computational ecosystems, including those used in research and production environments.
Technical Evolution
The transition to Keras 3 marked a significant architectural shift, enabling users to write custom layers, models, or metrics that can run natively in JAX, TensorFlow, or PyTorch without modification. This cross-framework capability addressed fragmentation in the deep-learning tooling landscape. The library also integrated with Data Augmentation techniques and other utilities to support diverse workflows. Keras 2, the earlier version, remains functional for projects that rely on its specific behavior, ensuring backward compatibility for existing codebases.