A machine learning framework is a tool, library or interface which helps developers to build machine learning models quicker and more easily. They decrease the need for building and maintaining complex algorithms, as they provide developers with fairly simple access to them. The following presents a brief overview of the three most popular machine learning frameworks at the moment. Obviously, there are a lot more frameworks currently available, so this list merely presents a brief overview.
Currently the most used framework for machine learning, TensorFlow's applications range from discovering new planets (https://www.nbcnews.com/mach/video/nasa-s-kepler-telescope-discovered-a-new-exoplanet-with-google-s-help-1121785923978, retrieved on March 18th 2019) to medical applications, such as preventing blindness (https://www.wired.com/2016/11/googles-ai-reads-retinas-prevent-blindness-diabetics/, retrieved on March 18th 2019).
Facts:
Developer: Google Brain Team
Initial Release: November 9, 2015
License: Apache 2.0 open-source license
(https://en.wikipedia.org/wiki/TensorFlow, retrieved on March 18th 2019)
Opposed to TensorFlow, Keras is not a standalone machine learning framework, but rather an interface, that can be used with a variety of frameworks, including TensorFlow. Its popularity can be explained by its comparatively simple approach, making it an ideal starting point for those new to machine learning.
Facts:
Developer: Original author: François Chollet (Google engineer)
Initial Release: 27 March 2015
License: Open Source (MIT License)
(https://en.wikipedia.org/wiki/Keras, retrieved on March 18th 2019)
PyTorch is the second most used standalone machine learning framework. Compared to TensorFlow, it allows for more customization and was established more recently. It has quickly gained popularity, most likely due to its flexibility.
Facts:
Developer: Facebook's artificial-intelligence research group