What Is Google’s TensorFlow?
Take a look at Google’s example of machine learning.
While Google’s contribution to the modern internet is hard to overstate, it has also missed the mark in a number of key areas. Its failed social media offering survives only through the benevolence of senior executives, while Google Video was so unsuccessful in its battle with YouTube that Google eventually acquired YouTube outright, and reduced its own offering to a sub-branded search engine page.
Google’s major rivals – Microsoft and Amazon – have left Google behind in the development of cloud products. AWS and Azure have enjoyed a healthy head start as providers of compute-as-a-service infrastructure, so Google is now attempting to make up lost ground with the rollout of its eponymous Cloud Platform. One of the more interesting components of this service portfolio is TensorFlow – an open source software library that can represent numerical computation as dataflow graphs.
So, What Exactly Is TensorFlow?
Born from the investigative Google Brain Team, TensorFlow represents an example of machine learning. This is the process whereby computers construct and refine algorithms that can learn from historic data to make predictions about future trends and likely events. As an offshoot of computational statistics, examples of machine learning include spam filtering and predictive search engine queries – one area where Google is the undisputed world leader.
TensorFlow OpenSourced
Written in Python but also compatible with C and C++, TensorFlow can be installed onto Linux, Windows or OS X devices. It can run in its own environment or a Docker container, as well as being installed directly onto a hard drive. Flexibility is one of TensorFlow’s greatest attributes, though the platform is still evolving; the release of its source code for open source development is a tacit acknowledgement by Google that they are far from the leading authority in this niche sector.
Machine learning is one of the specialist disciplines that will support the Internet of Things, which will eventually see vast quantities of statistical data being uploaded to remote servers from billions of web-enabled devices. The amount of processing required to assimilate and store this information will necessitate scalable cloud-hosted solutions, while the ability to sift through reams of data and draw meaningful conclusions will become crucial to the success (or otherwise) of those devices. The IoT won’t revolutionize our lives, but it is expected to generate untold levels of personal information that will be fairly useless without computational analysis being carried out and fed back to us in meaningful forms.
What’s Next?
It could therefore be argued that TensorFlow has arrived at an optimal time in the development of tomorrow’s web-based services. Its flexible architecture allows clients to allocate resources to a variety of CPUs and GPUs in setups ranging from a server to a mobile device. Google’s own uses for the architecture behind TensorFlow include speech recognition, search engine predictions and interpreting photos – the science behind Google Goggles and its supposed ability to view an image and then call up related search results. While this represents another example of a largely unsuccessful service launch by the California software giant, the computational knowhow underpinning Google Goggles will be crucial in developing the next generation of machine learning infrastructure.