Tutorial : Scalable Deep Learning with Microsoft Cognitive Toolkit

Emad Barsoum, Sayan Pathak and Cha Zhang;
We will introduce Microsoft’s Cognitive
Toolkit, also known as CNTK, to the computer vision
community. CNTK was originally developed by the
speech team at Microsoft Research, and it was the key to
Microsoft Research’s recent breakthrough in reaching
human parity in conversational speech recognition. In
February 2016, CNTK was released on GitHub with MIT
license, and today’s 2.0 version contains both C++ and
Python APIs, and runs natively on both Windows and
Linux. While CNTK is little known to the vision
community, it is highly regarded by other communities,
achieving at least 5-10x speed up on recurrent neural
networks compared with any other toolkits, on either CPU
or GPU. In this tutorial, we will explain how this is done
with symbolic loops, automatic batching and innovative
data parallel training algorithms. Furthermore, we will use
many computer vision related examples to demonstrate
the flexibility, extensibility and scalability of CNTK.

[TimScarfe]

Amazing video but the sound goes dead after about 2:20, help!