Complex Spiking Neural Network Evaluated by Injury Resistance Under Stochastic Attacks

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Network Science

Background:

• Brain-inspired models are vital for AI but struggle with environmental complexities and lack bio-plausibility.
• Enhancing the injury resistance of artificial neural networks is crucial for robust AI systems.
• Human brains exhibit self-adaptive abilities under injury, offering a blueprint for resilient AI.

Purpose of the Study:

• To propose a novel brain-inspired model, the complex spiking neural network (Com-SNN), with enhanced bio-plausibility and injury resistance.
• To investigate the injury-resistance mechanisms within the Com-SNN, focusing on synaptic plasticity and network topology.
• To compare the injury resistance of the Com-SNN against alternative spiking neural networks (SNNs) under simulated attacks.

Main Methods:

• Developed the Com-SNN model with topology inspired by biological brain networks, utilizing Izhikevich neuron models and time-delayed synaptic plasticity.
• Evaluated Com-SNN injury resistance using two metrics and compared it with other SNNs under simulated neuron removal (stochastic attacks).
• Analyzed synaptic plasticity dynamics and topological characteristics of the Com-SNN during stochastic attacks to understand resilience mechanisms.

Main Results:

• The Com-SNN exhibited significantly superior injury resistance compared to SNNs with alternative topologies.
• Experimental results confirm the potential of the proposed model to improve the overall injury resistance of spiking neural networks.
• The study successfully demonstrated the enhanced resilience of the Com-SNN under simulated physical attacks.

Conclusions:

• Synaptic plasticity is identified as a fundamental factor contributing to the injury resistance of brain-inspired models.
• Network topology plays a critical role in determining the resilience and robustness of spiking neural networks.
• The findings provide insights into developing more robust and bio-plausible artificial intelligence systems.