TensorFlow is an open source software library for programming data streams across various tasks. It is a symbolic math library, and is also used for machine learning applications such as neural networks. It's used for research and production on Google,? often replacing its predecessor with closed source, DistBelief.
TensorFlow was developed by the Google Brain team for internal Google use. It was released under the open source license of Apache 2.0 on November 9, 2015.
Video TensorFlow
History
DistBelief
Beginning in 2011, Google Brain built DistBelief as a proprietary machine learning system based on deep-learning neural networks. Its use is growing rapidly across a variety of Alphabet companies in research and commercial applications. Google commissioned several computer scientists, including Jeff Dean, to simplify and update the DistBelief code base into a faster and more powerful classroom application library, which became TensorFlow. In 2009, a team led by Geoffrey Hinton, has implemented a common backpropagation and other improvements that enable the formation of neural networks with much higher accuracy, such as 25% reduction in voice recognition errors.
TensorFlow
TensorFlow is Google Brain's second generation system. Version 1.0.0 was released on February 11, 2017. While reference implementations run on a single device, TensorFlow can run on multiple CPUs and GPUs (with optional CUDA and SYCL extensions for general-purpose computing on graphics processing units). TensorFlow is available on 64-bit Linux, macOS, Windows, and mobile computing platforms including Android and iOS.
TensorFlow calculations are expressed as stateful dataflow graphs. The name TensorFlow comes from the operation of such neural networks on multidimensional data structures. This array is called a "tensor". In June 2016, Dean stated that 1,500 repositories in GitHub mention TensorFlow, which only 5 are from Google.
Tensor processing unit (TPU)
In May 2016, Google announced a Tensor processing unit (TPU), ASIC tailor-made for machine learning and tailored for TensorFlow. TPU is a programmable AI accelerator designed to provide high-precision low-arithmetic output (for example, 8-bit), and oriented to use or execute a model rather than train it. Google announced that it has been running TPUs in their data centers for over a year, and has found them to provide a higher order of magnitude optimized performance per watt for machine learning.
In May 2017, Google announced the second generation, as well as TPU availability on Google Compute Engine. The second generation TPU delivers up to 180 teraflops of performance, and when set into a 64 TPU cluster, it provides up to 11.5 petaflops.
In February 2018, Google announced that they made TPU available in beta on Google Cloud Platform.
TensorFlow Lite
In May 2017, Google announced a collection of special software for Android development, TensorFlow Lite, starting with Android Oreo.
Apps
Google officially released RankBrain on October 26, 2015, which is supported by TensorFlow.