Microsoft Cognitive Toolkit 2.0

Also, "Reasons to Switch from TensorFlow to CNTK": https://docs.microsoft.com/en-us/cognitive-toolkit/reasons-t...

The important feature is Keras compatability (although it doesn't seem to be in the official repo yet: https://github.com/fchollet/keras/pull/6800)

In terms of "how is CNTK better than TensorFlow," CNTK trains 5-10x faster at minimum on pre-2.0 LSTM benchmarks, which is big since LSTMs are used for a lot nowadays. (https://arxiv.org/abs/1608.07249)

1-bit SGD is cutting-edge deep learning tech. (so cutting edge it has a different license than CNTK itself: https://docs.microsoft.com/en-us/cognitive-toolkit/CNTK-1bit...)

"Many other toolkits create examples that appears to be identical to the original implementation of a paper, and then offer a converted model for people to download and evaluate. This is highly irresponsible in our opinion. In CNTK we pay a lot of attention in tracking down bugs and make sure the toolkit can be used to actually train the model from scratch and achieve state-of-the-art accuracy.

One example is the Inception V3 network (https://arxiv.org/abs/1512.00567) originally developed by a few researchers at Google. Tensorflow shared the training script for Inception V3, and offered pre-trained models to download. However, no one were able to retrain the model and achieve the same accuracy with their script. The best accuracy that were achieved by a third party (Keras in this case) is about 0.6% worse that what the original paper reported. Researchers in the CNTK team worked hard and were able to train a CNTK Inception V3 model with 5.972% top 5 error, even better than the original paper reported! The training script is shared as an example (https://github.com/Microsoft/CNTK/tree/master/Examples/Image...) and you may verify it yourself."

Ha ha, that does read like sour grapes.

Is the idea that they will somehow provoke comparison with the 800lb gorilla, even at the expense of looking juvenile? What a bunch of whiny cry babies!

They refer to this benchmark in the blog post - http://dlbench.comp.hkbu.edu.hk/

There is also the v7 benchmark done on a lot more hardware and where tensorflow fares a bit better - http://dlbench.comp.hkbu.edu.hk/?v=v7

Does anyone know whether TF had a performance regression between v0.11 and v1.0 or if it was just lucky on benchmark v7 and unlucky on v8?

Also, how does CNTK manage to be that much better than anyone else on LSTMs? It's ability to scale to bigger batch sizes is unreal. Order of magnitude faster than other frameworks.