The biological neural network capable of effectively distinguishing external sound signals is a creative achievement of the National Research University Saratov.
According to their perspective, this invention could significantly reduce energy consumption compared to conventional artificial neural networks. The research results have been published in the scientific journal Chaos.
Modern technology has widely adopted various signal recognition methods using second-generation artificial neural networks. However, the neurons used to model real neurons are much more complex than those of artificial neural networks. As a result, the biological neural networks (explosive growth) of the third generation are quite different from those of the second generation. In recent years, the scientific community has increasingly focused on studying explosive growth neural networks, yet many unanswered questions remain.
Illustration of artificial neural networks.
As it is known, the neural networks in the human brain comprise groups of neurons connected chemically or functionally. Experts describe their behavior using the conceptual mathematical model FitzHugh-Nagumo, proposed in the late 20th century.
Scientists from the National Research University Saratov named after N.G. Chernyshevsky have set themselves the task of using the FitzHugh-Nagumo conceptual mathematical modeling method to determine the capability of sound signal recognition through explosive growth neural networks consisting of neurons. They hypothesize that networks based on such neurons may possess greater capabilities due to their integrated complexity.
“We focus on studying how networks of such neurons operate in relation to external signals. The network under study is not large, but the number of its elements is sufficient to achieve the desired effect. We discovered that the FitzHugh-Nagumo neural connection can exhibit selectivity towards signals with different frequency rates and distinguish external signals by selectively choosing certain connections between neurons. From this, we conclude that it is possible to construct a network of specific neurons to ensure the recognition of sound signal fields,” explained Andrei Bukh, Associate Professor at the Department of Radiation Physics and Nonlinear Dynamics of the National Research University Saratov.
The researchers believe that their new invention will help create efficient neural networks, where efficiency should be understood as the ratio between energy consumption and the complexity of the task being solved.
“It is known that to solve the same problem, the human brain consumes less energy than conventional computers. This means that explosive growth neural networks may consume significantly less energy compared to conventional artificial neural networks,” emphasized the Saratov scientist.
He added that due to the non-linearity of the components of explosive growth neural networks, it will become very complex, and the responses of the neurons within it may vary greatly. Therefore, measuring efficiency will be quite challenging. In his view, this can only be realized when explosive growth neural networks begin to be applied practically.
“We are exploring the question of the connectivity capacity that ensures the selectivity of the network in relation to external signals for the simplest specific neurons and have found promising results. However, for each specific task, the level of efficiency gain will vary. There are other difficulties, and the most serious of these is that ‘only a very small number of methods can be applied to train explosive growth neural networks,'” noted Associate Professor Bukh.
As he noted, the results obtained are primarily guaranteed by selectively choosing certain connection links between the neurons. The remaining links will be disconnected. If all connections between the neurons are activated, the network will not exhibit selective properties. Conversely, an insufficient number of connections will lead to almost complete absence of responses within it.
In the future, researchers from the National Research University Saratov plan to explore whether a single neuron model can “accumulate” signals and demonstrate different behaviors depending on the “context.”
“Preliminary results indicate that the neuron model accumulates input signals. That is, the history of the neuron affects its current state, and it reacts to the ‘context.’ However, whether a network based on such neurons can become a classifier remains an open question,” concluded expert Andrei Bukh.
