CWI researchers Bojian Yin and Sander Bohté, along with Federico Corradi, a colleague of Stichting Interuniversitair Micro-Elektronica Centrum (IMEC) in Eindhoven, have achieved mathematical breakthroughs in the calculation of so-called spiking neural networks.
Thanks to this breakthrough technology, this special chip for artificial intelligence (AI) can recognize voice, gestures and electrocardiogram (ECG) 20,000 to 1000 times more efficiently than traditional AI technology. .. Such chips are at stake for practical everyday applications.
Research results were published in scientific journals Nature Machine Intelligence October 14, 2021.
Over the last decade, AI has acquired more and more everyday applications such as image and spoken language recognition.This is done deep neural network, This is a very simplified imitation of how the human brain processes information.for Mobile applicationHowever, running the current AI model is often too costly. energy.. Therefore, the development of low-power AI is becoming more and more important.
One way to increase the energy efficiency of AI applications is to bring neural networks closer to the networks of the human brain. Traditional neural networks use continuous, mathematically easy-to-process signals. Spiking neural networks calculate with pulses. This is very similar to what happens in the brain and requires less energy, but it has the disadvantage that the signals are discontinuous and difficult to process mathematically. However, Bohté and his two co-authors have found a mathematical solution to the problem.
“we Computer algorithm “With three benchmarks,” says Bohté. “These benchmarks consist of about 10 gestures, a series of words, and a test set of continuous ECG signals. Our algorithm works at least as well, but with much more energy than before. Higher efficiency Deep neural network.. Theoretically, it would be 100 to 1000 times. ”
Using algorithms like Bohté in everyday applications requires a special neuromorphic computer chip. The architecture of these chips is more similar to the biological architecture of the human brain than that of traditional computer chips.”Based on our algorithm, our research partner IMEC has created a special neuromorphic,” said Beaute. Tip 336 spiking neurons: μBrain chip. Running the algorithm on this special chip will increase energy consumption by a factor of 20. Compared to the theoretical energy gain, the actual energy gain is always lower due to the conversion of digital signals to analog signals and vice versa, and the reading of data. However, there are still many 20-fold energy gains. To detect heart defects, it means that you can implant an ECG recording chip and it will run on a single battery for a year. ”
Over the next few years, neuromorphic chips will contain more and more spiking neurons, further expanding the possibilities of artificial intelligence applications in wearable chips. For example, at the end of September, American chip maker Intel already produced the neuromorphic chip Loihi2, which contains one million spiking neurons.
Bojian Yin, Accurate and Efficient Time Domain Classification with Adaptive Spike Recurrent Neural Networks, Nature Machine Intelligence (2021). DOI: 10.1038 / s42256-021-00397-w.. www.nature.com/articles/s42256-021-00397-w
Centrum Wiskunde & Informatica
Quote: Energy-efficient AI is a cardiac defect obtained on October 15, 2021 from https: //medicalxpress.com/news/2021-10-energy-efficient-ai-heart-defects.html (2021) October 15th) will be detected
This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. The content is provided for informational purposes only.
Energy-efficient AI detects heart defects
Source link Energy-efficient AI detects heart defects