The growing area of deep learning spawns the creation of a new type of compilers, which are designed to squeeze the most out of hardware running various convolutional networks.
Using 3 examples – Intel® Nervana™ Graph (nGraph™), Facebook Glow and NVIDIA TensorRT™, I’d like my talk to be an introduction to this type of compilers. To better understand the topic of compilers itself I would be also providing a basic explanation of regular compiler.
This explanation would allow later extrapolate the knowledge on the topic of deep learning compilers. Using 3 mentioned state-of-the-art compilers I intend to present common building blocks as well as briefly describe significant differences among them.
As a final point, I’d be highlighting differences and similarities between deep learning and regular compilers.