Such is the growth in demand for high speed processing of compute intensive AI algorithms, that Graphcore has spent two years in the lab developing hardware that will accelerate model training by a factor of 10 to 100. The firm has just received $30 million series-A funding from the likes of Robert Bosch Venture Capital and the Samsung Catalyst Fund (and several others), and surprise, surprise, this is not a Silicon Valley company, but is located in Bristol, England.
Some initial applications for the hardware include natural language processing, autonomous vehicles and personalized medicine. The IPU (Intelligent Processing Unit) will come to market in 2017 in the form of an appliance, and can be deployed in cloud and enterprise datacenters. The company also plans to make its low power IPU technology available for embedded consumer applications including autonomous cars, collaborative robots and intelligent mobile devices.
IPU systems will accelerate the full range of training, inference, and prediction approaches. Its huge computational resources and software tools and libraries are flexible and easy to use, allowing researchers to explore machine intelligence across a much broader front than the current focus on feed-forward neural networks. This technology will enable recent success in deep learning to evolve rapidly towards useful, general artificial intelligence.
The focus on deep learning is significant, since this is perhaps the most promising area of AI, as it offers the most likely route to general AI, although this is some way off. Deep learning involves the generation of a special form of neural network that might implement thousands or millions of neurons, and powerful compute resources are needed to train the networks. Oddly enough no one really knows why deep learning is so effective.
Graphcore CEO and co-founder, Nigel Toon, said, “Machine intelligence will have a bigger impact on our lives over the next 10 years than mobile technology has had in the last two decades. Next generation machine intelligence will allow us to translate foreign languages in real-time, help diagnose illnesses and develop personalized treatments, control robots that clean our houses and offices, drive cars autonomously and provide us with intelligent digital assistants that can help us organize our busy lives. The IPU is the first system specifically designed for machine intelligence.”
More information can be found at https://www.graphcore.ai