Dendrites endow artificial neural networks with accurate, robust and parameter-efficient learning

Artificial neural networks (ANNs) form the basis of most successful Deep Learning (DL) algorithms$^1$, which are capable of solving complex problems such as image recognition and natural language processing$^{2,3}$. However, unlike biological brains, which efficiently solve similar problems, DL algorithms require a large number of adjustable parameters, making them energy-intensive and susceptible to overfitting. In this study, a new ANN architecture is introduced that incorporates the structured connectivity and restricted sampling properties of biological dendrites, aiming to overcome these limitations$^4$. The research reveals that dendritic ANNs are less prone to overfitting and outperform traditional ANNs in various image classification tasks, while also using significantly fewer adjustable parameters. This achievement is realized through the application of a distinct learning strategy where most nodes respond to multiple classes, unlike classical ANNs that pursue class-specific responses. These findings indicate that the integration of dendrites can enhance the precision, resilience, and parameter efficiency of learning in ANNs, offering insight into how biological features can influence the learning strategies of ANNs.