minds.ai inc. (pronounced: maɪndz-aɪ), developers of an innovative deep neural network training platform with dramatic acceleration performance, today introduced its industry-leading platform that equips a new generation of developers with the tools needed to quickly refine and deploy neural networks into their businesses.
Delivering significantly reduced training times, the minds.ai platform is currently training AlexNet (a well-known task used for benchmarking training performance) with wall clock speeds that are four times faster than the fastest known publicly available benchmark1 (three hours compared to 12 hours to achieve 50 percent accuracy). minds.ai is using seven NVIDIA GeForce GTX TITAN X graphics cards (GPUs) as a proof of concept to demonstrate the performance of its parallelization algorithms. The minds.ai acceleration platform is the first to market with the unique combination of state-of-the-art speed and low barrier to entry from a usability and cost perspective.
“Neural network development is hamstrung by slow training speeds. We set out to build the fastest and most efficient training system today,” said Sumit Sanyal, CEO and Founder of minds.ai. “A 400 percent speedup in wall clock time means that a training run that used to take a month will now take a week. In the near future, our platform will scale up to hundreds of GPUs, 4x is just the beginning.”
The minds.ai platform is a combination of hardware and software. minds.ai has leveraged its team of neural network designers and supercomputing experts to develop proprietary training algorithms that have been designed to run on specific High Performance Compute hardware configurations.
As the minds.ai platform matures in 2016, beta testers will be selected to run training jobs in exchange for user feedback. Interested parties are encouraged to sign up at minds.ai for consideration in the developer beta testing program.
Source : minds.ai
with Business Wire
[1] http://yahoohadoop.tumblr.com/post/129872361846/large-scale-distributed-deep-learning-on-hadoop
© HPC Today 2024 - All rights reserved.
Thank you for reading HPC Today.
