TensorFlow is a software library for machine learning and artificial intelligence, used mainly to train and run neural networks. It supports a wide range of applications across many sectors through APIs in Python, JavaScript, C++, and Java.
Who created TensorFlow and where did it come from?
TensorFlow was built by the Google Brain team, evolving out of an earlier proprietary system called DistBelief that Google had used internally since 2011. Computer scientists including Jeff Dean were assigned to rebuild that codebase into TensorFlow.
When was TensorFlow first released?
The initial version of TensorFlow was released under the Apache License 2.0 in 2015, with version 1.0.0 following on the 11th of February 2017. Google released the major update TensorFlow 2.0 in September 2019.
What is a Tensor Processing Unit (TPU) and how does it relate to TensorFlow?
A Tensor Processing Unit, or TPU, is an application specific integrated circuit that Google announced in May 2016, built specifically for machine learning and tailored to TensorFlow. It is designed for high throughput, low precision arithmetic and is oriented toward running models rather than training them.
What is the difference between TensorFlow and LiteRT (formerly TensorFlow Lite)?
TensorFlow is the core library for training and running machine learning models on desktops and servers, while LiteRT, renamed from TensorFlow Lite in 2024, provides APIs for deploying compressed models on mobile apps and embedded devices. LiteRT uses the FlatBuffers format instead of the Protocol Buffers format used by standard TensorFlow models.
What companies and platforms use TensorFlow in real applications?
GE Healthcare has used TensorFlow to improve MRI speed and accuracy, and Google built the DermAssist skin app on it. Twitter has used TensorFlow to rank tweets, and Google released the RankBrain search system, backed by TensorFlow, on the 26th of October 2015.