Giving AI Roots to Grow: Cambridge Consultants Build Digital Greenhouse

by Emily Bryce

Interest in AI, and what it can achieve, is exploding.

But with growing hype around AI, it can be hard to discern how to best make use of the technology.

To shine a light on the situation, Cambridge Consultants, a global product development and technology consulting firm, introduced Digital Greenhouse. This initiative provides an open space for their engineers, developers and scientists to grow their understanding of AI using the NVIDIA DGX-1 AI supercomputer.

As avid gardeners know, it takes dedication, knowledge and hard work to grow anything. These are necessary if AI is to flourish.

Cambridge Consultants created its Digital Greenhouse several years ago as a place where researchers can toil and give AI roots a chance to grow. It enables them to push the limits of their understanding — outside of client projects — and share results in a spirit of openness and even fun.

Monty Barlow, technology director of Machine Learning at the company, explains: “The Digital Greenhouse is an experimental approach that engages the spare time and the creative inspiration of our smartest minds. It sees our technologists fired up and creative, knowing that they’re working at the frontier of an emerging, transformative industry.”

Having a good set of tools also helps. The NVIDIA DGX-1 allows Cambridge Consultants’ brightest minds to explore the possibilities of AI.

“The DGX-1 is reliable and constantly available,” says Barlow. “It is capable of working hard and continuously — we have the DGX-1 running 24 hours a day, 7 days a week.”

Transparency is Key

Just like a physical greenhouse with its glass walls, Cambridge Consultants’ initiative is transparent.

The company hopes that by making much of its work public, it will show what is possible with AI, and debunk some of the myths surrounding the technology.

This is why the Digital Greenhouse is not a shelter for only one species of AI, but a place of experimental growth for as many strains as possible, from generative adversarial networks to reinforcement learning.

For example, the Digital Greenhouse has shown that a deep learning system trained on 200 hours of four genres of piano music can categorize in real time previously unheard pieces better than any hand-coded approach.

Researchers have also been able to demonstrate the power of reinforcement learning through the classic arcade game Pac-Man:

AI Podcast: On the Origin of GANs

If you’re interested in finding out more about the issues surrounding AI and its development, be sure to check out the AI Podcast.

In the episode below, Ian Goodfellow, a staff research scientist at Google, explains how he got the idea for generative adversarial networks.