Accelerating Neural Networks

Lots of work underway in making nets faster, and extending their architecture.

DarwinAI Generative Synthesis Platform and Intel Optimizations for TensorFlow Accelerate Neural Networks in GlobalNewswire

DarwinAI-Intel Combination Delivers 16.3X Speedup on Image Classification Networks, Additional Inference Performance Improvements

WATERLOO, Ontario, July 10, 2019 (GLOBE NEWSWIRE) -- DarwinAI, a Waterloo, Canada startup creating next-generation technologies for Artificial Intelligence development, today announced that the company’s Generative Synthesis platform – when used with Intel technology and optimizations – generated neural networks with a 16.3X improvement in image classification inference performance. Intel shared the optimization results in a recently published solution brief.

“The complexity of deep neural networks makes them a challenge to build, run and use, especially in edge-based scenarios such as autonomous vehicles and mobile devices where power and computational resources are limited,” said Sheldon Fernandez, CEO of DarwinAI.  “Our Generative Synthesis platform is a key technology in enabling AI at the edge – a fact bolstered and validated by Intel’s solution brief.”

DarwinAI and Intel Optimizations Deliver As Much As 16.3X Performance Increase

Intel engineers ran image classification performance tests with ResNet50 and NASNet, two popular neural networks, using Intel Optimizations for TensorFlow with Intel Math Kernel Library (Intel MKL and Intel MKL-DNN). The results: for ResetNet50, Darwin’s Generative Synthesis platform combined with Intel’s optimizations delivered a 16.3X improvement in inference speed over baseline measurements on an Intel Xeon Platinum 8153 processor.  Meanwhile, for NASNet, the DarwinAI-Intel combination yielded a 9.6X improvement in inference speed on the same hardware. .... "