Specialists from the Saratov National Research University named after N. G. Chernyshevsky, together with colleagues from the Berlin Technical University, have made an important discovery in the field of machine learning. Researchers have found a way to replace complex neural networks with oscillators with delayed feedback. The discovery of scientists can significantly simplify and reduce the cost of creating devices capable of performing artificial intelligence tasks in real time.
Oscillators are systems that can oscillate, like a pendulum or an electrical circuit. Adding delayed feedback, in which the system responds to its previous states, makes their behavior more complex. Such oscillators can mimic the dynamics of entire networks consisting of many interacting elements.
A group of scientists led by Vladimir Semenov, head of the computer modeling laboratory in electronics, demonstrated that two coupled bistable oscillators with delay can reproduce the behavior of a complex two-layer neural network.
In the course of research, specialists studied two key phenomena: stochastic resonance, in which noise amplifies a useful signal, and the propagation of a wave front, when one of the states of the system fills the entire space. The researchers first simulated the behavior of oscillators on a computer, and then created physical prototypes using electronic components. The results of the experiments fully confirmed the theoretical calculations.
Currently, the university continues to work on the application of such systems for modeling "spiking neural networks", which largely mimic the electrical activity of brain cells.
Read more on the topic:
Neural networks in Moscow hospitals have been trained to detect diseases in 39 areas
A neural network for diagnosing tumors was created in Russia
Neural network trained to determine a person's place of birth by immunity
