Strange Loop

Towards new deep learning abstractions on top of exist frameworks, underlying principles and distributed execution

The Nervana Graph (ngraph) is a Python library for implementing programs that convert descriptions of neural networks into programs that run efficiently on a variety of platforms. In designing ngraph we kept three guiding motivations in mind:

A modular and flexible library designed around a unifying computational graph to empower users with composable deep learning abstractions. Execution of these models with maximal computational efficiency without worrying about details such as kernel fusion/compounding or data layout. Enabling all of this on any user's hardware whether one or multiple CPUs, GPUs, and/or Nervana Engines. To achieve these goals, the ngraph library has three layers:

An API for creating computational ngraphs. Two higher level frontend APIs (TensorFlow and Neon) utilizing the ngraph API for common deep learning workflows. A transformer API for compiling these graphs and executing them on GPUs and CPUs.

Tristan Webb

Intel

Tristan Webb received his Ph.D. in Complexity Science from the University of Warwick in 2013. His contributions include neural network simulations of brain decision making invoking biologically derived parameters. Machine learning was an important part of his scientific workflow, providing analysis for the computational neuroscience simulations. After moving to San Diego he worked as a postdoc at USCD and later two years in the software industry specializing in Functional Programming. Tristan is a strong believer that Deep Learning is a major growth field, and he works as at Intel Nervana to help data scientists express deep learning graphical models