Microsoft was able to squeeze their deep-learning algorithms onto an RPi 3 in order to bring intelligence to small devices.
Love it or fear it, AI is advancing, and it’s coming to small/portable electronic devices thanks to advanced developments made by Microsoft. The software giant was recently successful at loading their deep-learning algorithms onto a Raspberry Pi 3 SBC. The advancement will obviously be a boon for anything, and everything IoT, which is on track to take the world by storm and speculation suggests there will be 46-billion connected devices by 2021- depending on whom you ask.
Regardless, Microsoft’s latest breakthrough will allow engineers the opportunity to bring about intelligent medical implants, appliances, sensor systems and much more without the need for incredible computer horsepower. Most AI platforms today utilize the cloud for all their hardware endeavors, certainly so with infant platforms such as Amazon’s Alexa and Apple’s Siri but Microsoft’s breakthrough will make those systems obsolete and unnecessary.
Microsoft is developing AI platforms that will be squeezed into hardware no bigger than this chip. (Image credit Microsoft)
To further put Microsoft’s development into perspective- the team is capable of taking algorithms that normally run on 64 and 32-bit systems and drop the requirements down to a single bit in some cases. What’s astounding is how this new development came about- all due to a flower garden. Ofer Dekel, Manager of Machine Learning and Optimization at Microsoft’s Research Lab in Redmond, Washington, needed a way to keep squirrels from eating his flower bulbs and birdseed, leading him to develop a computer-vision platform utilizing an inexpensive RPi 3 to alert him when there was an intrusion.
When the alert is triggered, the platform engages a sprinkler system to shoo away the culprits- an ingenious solution indeed. “Every hobbyist who owns a Raspberry Pi should be able to do that, today very few of them can,” stated Dekel. Yet, the breakthrough will allow just that and can even be installed on a tiny Cortex-MO chip like the one pictured above.
To get the deep-learning algorithms compressed enough to fit on the RPi 3 using just a few bits, Ofer and his team employed a technique known as sparsification, a technique that shave’s off unneeded redundancies. Doing so allowed them to devise an image detection system that could process 20-times faster on limited hardware without losing any accuracy. Still, the team hasn’t yet figured out a way to take ultra-sophisticated AI or a deep-neural network and compress it enough to fit on limited, low-powered hardware. Regardless, this is an unprecedented first step in doing so, and we can certainly expect advancements that will get us there sometime in not too distant future.
I'm working on some Pi projects at the moment. Instead of IoT projects... maybe I should be looking into AI.
Have a story tip? Message me at: cabe(at)element14(dot)com