Microsoft: Embedded Learning Library

Embedded Learning Library is the library developed by Microsoft Research, will allow to work with artificial intelligence on small low-power computers without permanent connection to the Internet.

2017: Publication

On August 23 the Microsoft corporation published the Embedded Learning Library designed to help to implement the systems of machine learning in computers with tiny processors, such as Raspberry Pi. They will not need permanent connection to a cloud and the Internet, saving all the computing opportunities offline, also they will be very difficult to be cracked, assure of Microsoft. The beta of library is already available on GitHub.

According to representatives of corporation, scope of similar devices is quite wide. For example, the "smart gloves" capable to distinguish and sound a sign language, sensors of humidity of the soil and even the brain implants warning the carrier about possible spasms. Similar technologies will find application in the most different scenarios, for example, in predictive service for identification and elimination of breakdowns even before their emergence.

The idea of the project came up at the head of machine learning and optimization of Microsoft Research Ofer Dekel when his garden was attacked by squirrels. Being a specialist in the field of IT, he solved a problem by means of technologies. Dekel wrote an algorithm for recognition of squirrels and started it on Raspberry Pi 3. As a result he received the observation system behind the backyard including the sprayer at appearance of wreckers.

Before one of the main barriers there was high cost and impracticality of devices for cloud data processing. We can allocate with a large number of opportunities the processor, smaller by the size, without performance penalty — Ofer Dekel said.

The smallest device on which the library was tested is the single board computer Arduino Uno having 2 kilobytes of RAM. The next step — writing of an algorithm for work of the systems of machine learning on Cortex M0 processors, of the size of a bread crumb.