When was TensorFlow launched?

Google released TensorFlow as open source on 9 November 2015, under the Apache 2.0 license. TensorFlow is a library for numerical computation and machine learning that represents a model as a dataflow graph: nodes are operations and edges carry tensors, multi-dimensional arrays, which is where the name comes from.

The Google Brain team built it as a second-generation system to replace DistBelief, which Google had used internally since 2011 but which was hard to configure and tied to Google's infrastructure. Jeff Dean and Rajat Monga's launch post said TensorFlow was twice as fast as DistBelief on some benchmarks, and a white paper by Martín Abadi and colleagues came out the same day. Version 1.0, with a stable Python API, followed on 11 February 2017.

If an interviewer probes the word launched, the precise answer is that Google used TensorFlow internally before the 2015 open-source release. Related dates are easy to confuse: Keras appeared separately in March 2015, and Google announced its TPU chips in May 2016.