Researchers at the University of Utah in the United States have presented a model that puts artificial intelligence (AI) into any device. This does not require complex calculations.
The new model will allow researchers to integrate artificial intelligence into household appliances that use embedded technologies to interact with each other or with the external environment (the “Internet of Things”). According to preliminary data, the devices improve data security and energy efficiency.
The MCUNet system designs compact neural networks that, despite limited memory and processing power, provide high speed and accuracy for deep learning of IoT devices.
Using MCUNet, scientists are coding two components needed for “tiny deep learning” – the work of neural networks on microcontrollers. One of the components is TinyEngine, an engine that works with resources, it resembles a simple operating system. TinyEngine is optimized to work with a specific neural network structure.
Existing neural architecture search methods for items start with a large pool of possible network structures based on a predetermined pattern. Then, researchers gradually find one with high accuracy and low cost. While the methods work, they are not the most efficient for tiny devices.
“We have a variety of microcontrollers with different power and memory sizes,” the researchers explain. “Therefore, we developed an algorithm [TinyNAS] to optimize the search space for various microcontrollers.” The customizable nature of the TinyNAS means that it can generate compact neural networks with the best performance for a particular microcontroller, without even having to tweak unnecessary parameters.