Microsoft expects to transfer artificial intelligence technologies on the Raspberry Pi and other micro
The team of developers at Microsoft put into life the ideas about changing the paradigm of artificial intelligence, made by the head of the company at the recent Build conference 2017. The essence of the changes is to transition from obliczeniach concept for the combination of cloud and smart «front edge».
Giving the device forming the boundary of the electronic world, artificial intelligence, you can eliminate their dependence on network connections, reduce traffic, reduce energy consumption.
As an example, the source gives the solution based on the Raspberry Pi 3, trained to recognize the protein and to deter them from the flower buds and feeders for birds. The scope of application of MEMS with artificial intelligence is very wide — from the sensors of the Internet of things and wearable electronics to medical devices.
A feature of these devices is the presence of very modest computing resources and small amount of memory.
To adapt the algorithms of artificial intelligence to these hardware capabilities, the researchers used two approaches: developing compression algorithms machine learning models, allowing you to transfer them to the micro, and create a set of highly specialized algorithms, optimal for a specific task. Allegedly, this allows to reduce the requirements of hundreds and thousands of times.
The main difficulty is to preserve the accuracy of the algorithms at a level sufficient for practical use.
Interim results of the work of a team of Microsoft experts are available on GitHub. The developers hope that in the future artificial intelligence tools will be widely available to all enthusiasts who use their Raspberry Pi projects and other similar systems.