Title: Modernizing the Standard Neuron Model in Artificial Neural Networks

Abstract:

Since their introduction in the 1950s, artificial neural networks (ANNs) have relied on the point neuron model, a framework widely used in neuroscience at the time, under the assumption that this biological analogy would enhance the emulation of brain function. However, extensive research in neuroscience has since revealed that the point neuron model is overly simplistic and fails to accurately capture many essential neural processes. Despite these findings, the standard neuron architecture within ANNs has remained unchanged. In this study, we replace the traditional model with a recently developed representation of cortical cells. Through both theoretical analysis and experimental validation, we demonstrate that adopting this more biologically realistic neural unit yields significant benefits without increasing the parameter count. These advantages include enhanced expressivity and robustness, faster learning speeds, and reduced requirements for memorization and training data.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC