Microsoft successfully integrated a micro-AI system into Rasberry Pi 3

Rasberry Pi 3 is one of the second most popular devices today. With only USD 35 (equivalent to VND 770,000), you can own a computer with a variety of applications, from PCs running Linux, to making compact game consoles, and many more great things. half. Now, Microsoft has announced that it has built a device from Rasberry Pi 3 capable of using artificial intelligence to catch nasty rodents: squirrels.

The device from Rasberry Pi 3 is capable of using artificial intelligence to capture squirrels.

This device was created by the director of Machine Learning of Microsoft, Mr. Ofer Dekel. He discovered that the squirrels were the culprits who stole flower bulbs with the seeds for birds on his garden's nest. Due to his age and lack of free time, Dekel could not spend the whole day watching and hunting those hateful mice. But he did not give up and resigned to this situation, so he came up with a plan.

Working with their staff at the research lab in Redmond, Washington, they trained a computer vision device to detect squirrels. This artificial intelligence is integrated into a Rasberry Pi 3 containing machine that he puts in his backyard. When a rat or squirrel runs through, the device will detect what it is and turn on the sprinkler system, hinder theft and teach rodents a lesson.

Super tiny chip placed on the coin.

This project is only a part of the future perspective that Microsoft poses.'We are in the process of moving from mobile devices, cloud data to a new world flooded with technology of artificial intelligence,' said Microsoft CEO Satya Nadella during the event for broadcasters. Developed Build recently.

The biggest achievement of the demineralization project , according to Microsoft, is 'stuffing' an artificial intelligence system into a very small size chip. Dekel and his team have 'used a lot of technology' to compress this system, which has a method called quantization , capable of 'cramming' multiple parameters into a small capacity. This compression makes the artificial intelligence system work at a faster rate.

The group of scientists created the micro-AI system.

In addition, Dekel and colleagues are also testing a technique called 'cropping' , which will help eliminate unnecessary components in the neural network. This method offers two benefits: being able to run the AI ​​system on a microchip, being able to handle processing many times faster.

However, the team's goal is to integrate artificial intelligence on the smallest ARM processor : Cortex M0 . According to ARM, the chip is only 0.007 square meters in size. In other words, it is very, very, very small and will need Microsoft geniuses to shrink the AI ​​model 10,000 times smaller than they did on Rasberry Pi 3.

This chip has an area of ​​only 0.007mm square.

'How hard is it that both make the deep neural network work correctly, and have the size of' lash 'and consume 10,000 times less energy. This is almost impossible , 'Dekel said. 'So we will implement a new approach, which is to start from the beginning. Rewrite the mathematical processes onto the whiteboard and invent new technologies and tools related to machine learning to create these groundbreaking platforms in the future. "