It is an open-source software library that provides a Python interface for artificial neural networks. 

It is possible to develop your own Keras models using various deep learning backends. The main feature is that any model using only embedded layers will be transferable from one backend to another.

Available backends:

  • TensorFlow (Google)
  • CNTK (Microsoft)
  • Theano

Keras is widely used in industry and within the research community.

Keras is one of the leaders among deep learning researchers, ranking second in the number of mentions in scientific articles uploaded to the server.

Clients:  CERN, NASA


Digital platforms: Cross-platform software

Versions: Cloud/On-Premise 


Keras is at the nexus of a large ecosystem of closely related projects that cover every stage of the machine learning workflow: 

  • Rapid model prototyping with AutoKeras;
  • Training a scalable model in GCP via TF Cloud;
  • Hyperparameter tuning with Keras Tuner;
  • Additional layers, losses, metrics, callbacks via TensorFlow Addons;
  • Quantization and chunking of the logical output model using the TF Model Optimization Toolkit;
  • Deploying the model on the mobile device using TF Lite.
  • Deploying the model in a browser via TF.js and more;

Models created in Keras can easily be deployed on more platforms than any other deep learning framework:

  • On iOS via Apple CoreML (Keras support is officially provided by Apple);
  • On Android via the TensorFlow runtime environment;
  • In the browser via the JavaScript graphical accelerator (Keras.js and WebDNN);
  • On Google Cloud via TensorFlow-Serving;
  • In a Python web application (e.g.: Flask application);
  • On JVM, via DL4J model import provided by SkyMind;
  • On Raspberry Pi.