Explore how the architecture of nGraph-TensorFlow bridge enables easy integration of hardware (such as CPU, GPU, and NNP) and software backend (Homomorphic Encryption and PlaidML). Walk through various features of nGraph using TensorFlow DL models and see how easy it is for end users to make minimal changes in their Tensorflow script to enable nGraph for optimizations and migration to new backends. Watch demonstrations of debugging tools, techniques to visualize the graph compilation process, and troubleshooting.
You Will Learn:
- How compilers designed specifically for deep learning can achieve significant performance increase even with the existing hardware.
- That deep learning compilers like nGraph & PlaidML are easy to use with TensorFlow.
- That these compilers require minimal changes to previously written TensorFlow script to execute.
Avijit Chakraborty is a Principal Engineer in the Artificial Intelligence Products Group at Intel Corporation. He currently leads the nGraph integration effort with TensorFlow*. Before joining Intel in 2017, he was a software architect and lead software development teams in deep learning framework, neuromorphic computing framework, and cellular modem technologies at Qualcomm Research and Development.