TensorFlow makes news in 2015

On the face of it, it would appear that TensorFlow received an inordinate amount of attention in 015 for just a machine learning engine. But publicity works that way. Google is a $66 billion-per-year company, and buzz automatically goes with that. But a series of pieces by Cade Metz in Wired were fairly illuminating.

You push the little valve down, and the music goes around, and it comes out.

It seems that Google open sourcing, to some extent, TensorFlow gave it some lift. My take would be that they would like others to write the Java and JavaScript Notebooks, and create long lists of libraries, to give it the panache of Apache Spark, which is provably hot, due to its metered Apache klingonage. Is it an attempt to take wind out of Spark's sails? Some people say that Spark is general-purpose, and thus not as good for machine learning as is TensorFlow, which has no other use in life but to do machine learning.

Yet another Wired piece by Metz discusses the upswing in use of GPUs for machine learning. One banner that TensorFlow seems to forward is the use of GPUs for machine learning. It seems to be a growing meme.

One wag opines that TensorFlow smells like something that might work autonomous car data. In the back of my mind I can hear the words of someone who told me the difference between IBM and Watson and Google and TensorFlow is that the former is about software to enable enterprises to make consumer products, while the latter is about making consumer products. [Just last week, Google said it would launch such a vehicle with Ford. Which could play to the idea that it doesnt want to make cars, it wants to get data on drivers.]

As described Metz's article deep learning is the same as machine learning. Metz points out that the new goop in the secret sauce is the increase in both available processing and available data – suggesting that the algorithms are not dramatically different than in the past. [Although the author goes on to say the algorithms are evolving, and that gifted individuals are behind that evolution.]

Author Metz and source Lukas Biewald of Crowdflower note that Google open sourced the machine learning software, but not the data. Others criticize the fact that, while you can run Tensorflow on your own machine (that could include a GPU board), they are keeping the distributed version to themselves.

Links
https://www.youtube.com/watch?v=ENZoY4mLgDE
http://www.wired.com/2015/11/google-open-sources-its-artificial-intelligence-engine/
http://www.kdnuggets.com/2015/11/google-tensorflow-deep-learning-disappoints.html
http://www.wired.com/2015/11/google-open-sourcing-tensorflow-shows-ais-future-is-data-not-code/


http://www.wired.com/2015/11/googles-open-source-ai-tensorflow-signals-fast-changing-hardware-world/
https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEi-5K0GDYQsrw_oRhC_2hQ0lOp5xC0n1GRxtnpViaFNmriieL3FWkHQRf9QJS_5W7ZMPilxA-MxibJO4A9dygaRSqj0x15klMEstjZ7JDbVtg282JnG8IRd4wTTBEfS828paM2UEQi-Vk4/s1600/cifar10_2.gif
http://www.tensorflow.org/
http://googleresearch.blogspot.com/2015/11/tensorflow-googles-latest-machine_9.html

http://googleresearch.blogspot.com/2015/11/computer-respond-to-this-email.html
https://www.youtube.com/watch?v=46Jzu-xWIBk
https://www.youtube.com/watch?v=gY9DewL6Dqk
http://blogs.nvidia.com/blog/2015/03/18/google-gpu/