CNTK is optimized platform for RNN ASR (automatic speech recognition). In terms of training and evaluation speed, CNTK has significant advantages over TensorFlow. A comparison made by researchers at HKBU (http://dlbench.comp.hkbu.edu.hk/, and article: https://arxiv.org/pdf/1608.07249.pdf)
For GPU trainning, CNTK would have a linear speedup according to mini-batch size, thus would be 10 times faster than other deep learning platforms ( theano, tensorflow or torch).
CNTK is optimized platform for RNN ASR (automatic speech recognition). In terms of training and evaluation speed, CNTK has significant advantages over TensorFlow. A comparison made by researchers at HKBU (http://dlbench.comp.hkbu.edu.hk/, and article: https://arxiv.org/pdf/1608.07249.pdf) For GPU trainning, CNTK would have a linear speedup according to mini-batch size, thus would be 10 times faster than other deep learning platforms ( theano, tensorflow or torch).
Do you have any plan on CNTK?