How to use artificial intelligence to optimize edge IoT

As more companies combine Internet of Things (IoT) devices with edge computing capabilities, people are increasingly curious about how artificial intelligence (AI) can be used to optimize these applications. Here are some thought-provoking possibilities.

Using machine learning to improve IoT sensor inference accuracy

Technology researchers are still in the early stages of investigating how to improve the performance of edge-deployed IoT sensors through machine learning. . Early applications include using the sensor for image classification or tasks involving natural language processing. But there is an example of how people can make progress.

Researchers at IMDEA Network are aware that if IoT sensors are used for specific deep learning tasks, it may result in the sensor being unable to guarantee a specific quality of service, such as experiencing delays and reduced inference accuracy. However, researchers involved in the project developed a machine learning algorithm called AMR2 to address this challenge.

AMR2 leverages edge computing infrastructure to make IoT sensor inferences more accurate while enabling rapid response and real-time analysis. Experiments show that compared with the results of basic scheduling tasks without using the algorithm, the reasoning accuracy after using the algorithm is improved by 40%.

They found that efficient scheduling algorithms like this are critical to helping IoT sensors work properly when deployed at the edge. One project researcher pointed out that if developers use the AMR2 algorithm for a service similar to Google Images (that classifies images based on the elements they contain), it may affect execution latency. Developers can deploy this algorithm to ensure that users do not notice such delays when using the application.

Edge Artificial Intelligence Reduces Energy Consumption of Connected Devices

A 2023 study of tech company CFOs showed that 80% of companies expected to see revenue increase in the coming year. But increasing revenue requires employees to understand customer needs and deliver products or services accordingly.

Many manufacturers of IoT devices want people to wear their products regularly. Some wearable devices can detect when a lone worker has fallen or is in pain; they can also detect when a physically demanding role is overtired and in need of a break. In this case, users must have confidence in their IoT devices that they will work reliably at work and beyond.

That’s one reason researchers are exploring how edge AI can improve the energy efficiency of IoT devices. IoT devices are used to study the impact of prolonged sitting on health and how correct posture can improve outcomes. Any IoT device that captures lifestyle data must collect data continuously, so there is little or no chance that it will stop collecting information because the device runs out of battery.

In order to avoid the above situation, the wireless devices worn by the subjects are usually powered by button batteries. Typically, every gadget has inertial sensors that collect accurate data on how much people move throughout the day. The main problem, however, is that due to the large amount of data being transferred, the battery power only lasts a few hours. For example, research shows that a nine-channel motion sensor that reads 50 samples per second will generate more than 100MB of data in a day.

However, researchers realized that machine learning could allow algorithms to transmit only critical data from IoT devices deployed at the edge to smartphones or other devices that help analyze the information. They continued using pre-trained recurrent neural networks and found that the algorithm achieved real-time performance and was able to improve the functionality of IoT devices.

Creating opportunities for device-side artificial intelligence training

Advances in edge computing provide opportunities to use smart devices in more places. For example, it has been proposed to deploy smart streetlights that can be turned on and off based on real-time traffic conditions. Technology researchers and enthusiasts are also interested in the increased training opportunities for artificial intelligence deployed directly on IoT devices at the edge. This approach can improve product functionality while reducing energy consumption and improving privacy protection.

A team at MIT has investigated the feasibility of training artificial intelligence algorithms on smart edge devices. They tried optimizing several techniques, one of which requires only 157K of memory to train machine learning algorithms on a microcontroller, while other lightweight training methods typically require 300-600 MB of memory. This innovation resulted in significant improvements.

Any data generated during training remains on the device, reducing the risk of privacy breaches, the researchers explained. They also present use cases for training during normal use, such as whether the algorithm can learn from typing on a smart keyboard.

This approach certainly achieved impressive results. In one case, the team trained the algorithm for just 10 minutes before it was able to detect people in images. This example shows that optimization can go both ways.

While the first two examples focus on improving the way IoT devices work, this approach also enhances the AI ​​training process. However, it would benefit both the AI ​​algorithms and the IoT edge devices if developers could train the algorithms on IoT devices and achieve better performance.

How to use artificial intelligence to improve the way IoT edge devices work?

These examples illustrate the focus of researchers as they explore how artificial intelligence can improve the functionality of IoT devices deployed at the edge. I hope these provide you with valuable insights and inspiration. It’s always best to start with a well-defined problem and then look for technologies and innovative approaches that can help achieve your goals.

The above is the detailed content of How to use artificial intelligence to optimize edge IoT. For more information, please follow other related articles on the PHP Chinese website!