What is it about?
Artificial neural networks are widely used in modern technologies, but they often rely on simple mathematical functions that do not fully reflect how real brain cells behave. In this work, we introduce a new approach by using Bessel functions—special mathematical functions known for their oscillatory and flexible behavior—inside neurons with multiple dendrites. These multi-dendritic neurons are closer to biological neurons, as they can process multiple inputs in a more dynamic way. By combining this structure with Bessel functions, we show that neural networks can learn faster, adapt better to complex data, and improve their ability to recognize patterns. This approach is particularly useful for analyzing time-varying signals such as brain activity (EEG), where capturing complex patterns is essential. Our results open new possibilities for improving artificial intelligence systems and may contribute to better tools for understanding and detecting neurological disorders such as epilepsy.
Featured Image
Photo by Growtika on Unsplash
Why is it important?
What makes this work unique is the introduction of Bessel functions as activation functions in multi-dendritic neuron models an approach that moves beyond traditional functions such as ReLU and sigmoid. Unlike standard methods, Bessel functions provide controllable nonlinearity and oscillatory behavior, enabling neural networks to better capture complex and dynamic patterns. This study is also timely, as there is growing interest in developing more biologically realistic and adaptable artificial intelligence systems. By combining advanced mathematical functions with neuron structures inspired by real brain cells, this work offers a new direction for improving learning efficiency, robustness, and generalization. The potential impact is significant in fields that rely on complex signal analysis, such as biomedical engineering and neuroscience. In particular, this approach could enhance the analysis of brain signals (EEG) and support the development of more accurate tools for detecting neurological disorders like epilepsy.
Perspectives
Working on this article was an exciting opportunity to connect advanced mathematical theory with neuroscience-inspired models. I have always been interested in exploring how complex mathematical functions, such as Bessel functions, can better represent the dynamic behavior of real neurons, especially in systems with multiple dendrites. This work reflects my broader research vision of developing more realistic and efficient neural models that go beyond traditional approaches. It was particularly rewarding to see how combining mathematical flexibility with biologically inspired structures can lead to richer and more adaptive neural dynamics. I hope this study encourages researchers to explore unconventional activation functions and inspires new ideas at the intersection of mathematics, artificial intelligence, and neuroscience. Ultimately, I believe this direction can contribute to better understanding brain activity and improving tools for analyzing neurological signals.
dr kaouther SELMI
Universite de Monastir
Read the Original
This page is a summary of: Behavior of Bessel functions in neuron with multi-dendrites, AIP Advances, July 2025, American Institute of Physics,
DOI: 10.1063/6.0004485.
You can read the full text:
Contributors
The following have contributed to this page
