Local processing of sensor data

Artificial intelligence is often uses to analyze metadata,

but a group of German researchers are trying to process more limit information on-chip and in-situ by designing special hardware.

Researchers at the Fraunhofer Integrated Circuits Institute in Germany are working to design machine learning networks that provide everything that artificial intelligence needs on-site,

without having to exchange data with the cloud. sensor data

The researchers believe that achieving this goal will have many new uses, including the real-time evaluation of data provided by sensors. As such, there will be no delay in sending unnecessary data, and data processing speeds will increase.

In addition, the devices can adjust themselves by enabling local processing on the chip and the ability to self-train the system or sensors.

 Even such systems can be reconfigured to perform different tasks and have an embedded system capable of performing a variety of tasks. Much of the data sent through the IoT is wasted and wastes resources.

For example, a temperature sensor reads the data once every 5 minutes.

 As a result, data is generated and transmitted even when the ambient temperature has not changed.

While we only want to know when the temperature has changed and whether or not this temperature has gone beyond the normal range?

Proposed by German researchers for use in Predictive Sensors or Predictive Sensors,

they will be applied in the future with the development of the Industrial IoT (IIoT).

This intelligent latent system, called AifES, is based on the RISC-V architecture and is equipped with a special hardware accelerator.

 An important feature of such a system is that the learning process is performed on a chip rather than relying on the cloud or running on a computer, without the need for an Internet connection, and is only needed to submit analysis results to the Internet, and German researchers call it Decentralized AI.

 Because its purpose is not metadata processing. It can even be extended and this network of sensors is equipped with Swarming.

 This allows the sensors to interact with each other and work independently of a central information management network.

 Researchers believe it is possible to build a network of small, adaptable systems that divide tasks between themselves. Another feature of these decentralized and localized neural networks is that they are safer than cloud computing. Because data is processed on-site, not in the cloud.

Leave a Reply

Your email address will not be published. Required fields are marked *