South Korean researchers developed an artificial brain system which can learn by imitating the human brain and judge with its cognitive ability. The breakthrough is expected to implement artificial intelligence (AI) with low power consumption by applying to mobile and Internet of Things (IoT) devices.
The research team led by Professor Choi Sung-jin of Kookmin University and Professor Kim Sung-ho of Sejong University announced on March 23 that it has developed “neuron imitating element,” which works like neurons in the human brain using carbon nanotubes. Traditional computers digitally encode information in bits that can be in one of two states, 0 or 1. However, the neuron imitating element is an electronic element that can process and learn information with an analog system like the human brain by imitating the electronic properties of neurons.
The research team simulated how the element recognizes human’s actual handwriting in cursive letters by applying the learning algorithm of neural networks in charge of visual information processing in the human brain. As a result, it could remember and identify different script images of each person through 60,000 times of iterative learning.
Its recognition accuracy of human’s cursive script image patterns reached 80 percent by improving analogue performance characteristics by 10 times. With the decrease in leakage current, power consumption is also expected to be down to one-hundredth of its original state, according to the research team. In particular, it was hard to make the existing neuron imitating element uniform in large areas but the research team succeeded in making it uniform with high density in large areas using carbon nanotubes which have high performance and safety. In addition, it succeeded in realizing the high integrated circuit by fundamentally eliminating interferences between elements with the new three-terminal element structure. The team then completed an artificial brain system by devising and applying a learning algorithm that can imitate the visual imaging process of human.
Professor Kim Sung-ho said, “It is meaningful in that we opened up another way of AI technology as we made the hardware itself work like the human brain.”
The findings of the research was published on ACS Nano, an international journal dedicated to nanoscience and technology.