The world is passionate about artificial intelligence (AI), which can process huge amounts of data. However, modern AI systems based on artificial neural networks consume a lot of energy, especially when working with data in real time.
Scientists proposed a new approach to “machine intellect”. Instead of software for artificial neural networks, they developed a physical neural network at a hardware level, which works much more efficiently. These neural networks, created from silver nan pipelines, are able to learn how to recognize manuscript numbers in real time and remember the sequences of numbers. The results of the study were published in the journal Nature Communications together with colleagues from the University of Sydney and the University of California.
Using nanotechnologies, scientists have created networks of silver nan pipelines, thickness in the thousandth of the human hair. These nano pipes form a random network resembling the structure of neurons in our brains. Such networks respond to electrical signals, changing the method of transmitting electricity at the intersection points of nan pipelines, which is similar to the work of biological synapses.
The study shows that nano-based networks can be used for online learning. Unlike traditional machine learning, where the data is processed by packages, in the online approach the data is submitted to the system in a continuous stream. This method of teaching on fly, which is more effective than traditional package learning, requires less memory and energy.
In experiments, a network based on nan pipelines has successfully coped with the tasks of recognition and memorization of numbers, demonstrating the potential for emulating brain-like learning and memory.
In general, research in the field of neuromorphic nanobrus nets is just beginning, and new horizons of capabilities are opened before scientists.