TensorFlow is an open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them. Researchers and engineers of Google Brain Team within Google’s Machine Intelligence research organisation developed TensorFlow for the purpose of conducting machine learning and deep neural networks research. This system designed is general enough to be applicable in a wide variety of other sectors also. It is basically a library for creating computational graphs built by Google Research. It helps in making progress with everything from translating the language to early detection of skin cancer and many more features. Earlier Google launched “Google Translate” which work on the use of “Phrase-Based Machine Translation” as the core concept behind this service. As a result of this, it has helped in improving speech recognition and machine recognition capabilities of the machines, but still after this revolutionary feature, improving machine translation remains a challenging goal. Recurrent Neural Networks (RNNs) was launched to directly learn the mapping between an input sequence to an output sequence. Since then, many researchers have proposed many techniques to improve Neural Machine Translation (NMT). Despite many improvements, NMT wasn’t fast or accurate enough to compete with Google Translate. Today Google has launched Google Neural Machine Translation System (GNMT), which utilises state of the art of training techniques to achieve the largest improvements to date for machine translation quality.
With the launch of latest version TensorFlow 1.0, major changes have been done with the APIs with the effect of which many programs which were compatible with TensorFlow 0.N won’t necessary work on TensorFlow 1.0. This major step of API changes has been taken to ensure an internally-consistent API because of which many backwards breaking changes throughout the 1.N lifecycle will not be possible. TensorFlow also has the popular module Keras built right in to simplify and accelerate the deep learning development. Keras is a high-level neural networks API, written in Python and capable of running on top of either TensorFlow or Theano. It was developed with a focus on enabling fast experimentation.
TensorFlow provides well-tested, pre-built TensorFlow binaries for Linux, Mac, and Windows systems along with TensorFlow docker images. So no need to build a TensorFlow binary yourself unless you are very comfortable building complex packages from source and dealing with the inevitable aftermath should things not go exactly as documented. TensorFlow doesn’t officially support Windows but it can be installed using highly experimental “Bazel on Windows” or “TensorFlow CMake build”. You get to choose from two options you get while building and installing TensorFlow, these are:
- TensorFlow with CPU support only
- TensorFlow with GPU support
At the base of this upgradation, there is a new technology that enables to learn a rich layered representation of input sentences. This upgrade extends TensorFlow to allow joint modelling of multiple levels of linguistic structure and to allow neural-network architectures to be created dynamically during processing of a sentence or document.
Hope this article helped you understanding its basic idea. Follow to know more and keep yourself updated.
To learn more on TensorFlow Click HERE