Scientists at the Korea Advanced Institute of Science and Technology (KAIST) have developed a new neuromorphic chip that can learn and correct its own mistakes.
This development in artificial intelligence (AI) opens the way to creating more efficient and autonomous devices that can work without connecting to cloud servers. The research was published in Natural electronics.
The development of the team, led by professors Shinhyun Choi and Young-Gyu Yoon, is based on the use of a memristor, an element that imitates the work of synapses in the brain.
The memristor has a variable resistance, allowing it to store and process data at the same time, greatly increasing the efficiency of the chip.
The main problem with current neuromorphic devices is that they have imperfect properties that lead to errors. KAIST’s new chip can adapt and learn from these mistakes, improving accuracy over time.
This makes it ideal for use in video surveillance systems, medical devices and other applications that require immediate data processing.
The test results showed that the chip achieves an accuracy comparable to that of a perfect computer simulation, processing images in real time.
The scientists emphasized that the key to success was not just the creation of a separate component, but of an integrated system capable of solving complex problems in conditions of limited resources.
Due to their unique properties, memristor platforms enable the creation of compact and energy-efficient devices that can perform parallel computations in analog format. This significantly increases the speed and security of operation, reducing dependence on external servers.
As the researchers noted, the new technology changes the approach to implementing AI in everyday devices, providing the ability to process tasks directly on the spot, without the need to transmit data over the Internet.
This solution promises to make artificial intelligence faster, smarter and more accessible to a wide range of users.
