Biologically plausible unsupervised learning for self-organizing spiking neural networks with dendritic computation

Spiking neural networks (SNNs) are event-driven networks that possess efficient information processing capabilities. Compared to artificial neural networks (ANNs), SNNs are closer to the behavior of the brain in terms of information processing. Improving the biological plausibility of SNNs can help to explore and understand the computational mechanisms of the brain more effectively. In this paper, we propose a self-organizing SNN model with dendritic computation based on biological plausibility. Considering the impact of dendrites on neuronal and circuit functions, we proposed a multi-compartment neuron model with dendrites, which can enrich the dynamics and representation capabilities of neurons. Inspired by the organization of the brain network, we propose an adaptive self-organizing inhibition strategy that can cluster neurons encoding similar features together to help the network learn richer organizational relationships. In addition, we propose a partially shared connection method, which can effectively reduce the number of neurons in the network and significantly accelerate the training speed of the network. The experimental results on the MNIST, FashionMNIST, N-MNIST, and DVSGesture datasets show that our biologically plausible model has excellent potential to explore and understand brain mechanisms.