Deep Pulse-Coupled Neural Networks

TL;DR

This work introduces Deep Pulse-Coupled Neural Networks (DPCNNs) by substituting LIF neurons with biologically motivated PCNN units to enrich spatio-temporal dynamics for vision tasks. A key innovation is inter-channel coupling, enabling cross-channel communication, paired with receptive field–time dependent batch normalization (RFTD-BN) to accelerate training and improve generalization. Empirical results across MNIST, N-MNIST, Fashion-MNIST, and CIFAR-10 show DPCNNs achieving competitive or state-of-the-art accuracy with fewer neurons and train-time compared to widened LIF-based networks. The findings highlight the value of bio-plausible neuron models and configuration choices (inter-channel coupling, RFTD-BN, and STBP) for efficient, robust neuromorphic vision. Overall, DPCNNs advance brain-inspired AI by leveraging PCNN dynamics to enhance expressiveness while maintaining practical efficiency.

Abstract

Spiking Neural Networks (SNNs) capture the information processing mechanism of the brain by taking advantage of spiking neurons, such as the Leaky Integrate-and-Fire (LIF) model neuron, which incorporates temporal dynamics and transmits information via discrete and asynchronous spikes. However, the simplified biological properties of LIF ignore the neuronal coupling and dendritic structure of real neurons, which limits the spatio-temporal dynamics of neurons and thus reduce the expressive power of the resulting SNNs. In this work, we leverage a more biologically plausible neural model with complex dynamics, i.e., a pulse-coupled neural network (PCNN), to improve the expressiveness and recognition performance of SNNs for vision tasks. The PCNN is a type of cortical model capable of emulating the complex neuronal activities in the primary visual cortex. We construct deep pulse-coupled neural networks (DPCNNs) by replacing commonly used LIF neurons in SNNs with PCNN neurons. The intra-coupling in existing PCNN models limits the coupling between neurons only within channels. To address this limitation, we propose inter-channel coupling, which allows neurons in different feature maps to interact with each other. Experimental results show that inter-channel coupling can efficiently boost performance with fewer neurons, synapses, and less training time compared to widening the networks. For instance, compared to the LIF-based SNN with wide VGG9, DPCNN with VGG9 uses only 50%, 53%, and 73% of neurons, synapses, and training time, respectively. Furthermore, we propose receptive field and time dependent batch normalization (RFTD-BN) to speed up the convergence and performance of DPCNNs.

Figures (14)

• Figure 1: Illustration of the LIF and PCNN neuron model and their dynamics: (a) In comparison to the LIF neuron, the PCNN neuron takes into account the dendritic structure, which enhances its computational capabilities. (b) The LIF neuron, when subjected to constant stimuli, exhibits periodic firing patterns. (c) Conversely, a pair of coupled PCNN neurons under constant stimuli can generate bursts of spikes, demonstrating their ability to exhibit synchronized activity and complex dynamics.
• Figure 2: Diagram of the PCNN neuron.
• Figure 3: Diagram of the DPCNN model.
• Figure 4: comparison between intra- and inter-coupling.
• Figure 5: Computational gragh of DPCNN. Information propagates in both the spatial and temporal domains. The backpropagation path is not included in the diagram for the sake of brevity.