Russian Scientists Make Discovery That Could "Humanize" Neural Networks
Researchers in Saratov have uncovered how spiking neural networks can be tuned for stability and efficiency, bringing artificial intelligence closer to the way the human brain works.
Scientists at Saratov State University have made a discovery that could bring artificial intelligence technologies closer to the principles of the human brain—not just in design but in efficiency.
According to findings published in The European Physical Journal Special Topics, the focus is on spiking neural networks. Unlike traditional ones, they activate only when receiving a signal, making them more energy efficient.
By studying the effect of noise on such networks, the researchers observed the phenomenon of coherent resonance—when, at a certain level of interference, neuronal activity becomes ordered, boosting the stability of the system. The team concluded that efficiency depends on balance: neurons must be sufficiently connected to suppress noise while still able to transmit useful signals.
The research is expected to help refine next-generation neural networks that more accurately imitate the biological brain.
