Machine Unlearning in Brain-Inspired Neural Network Paradigms Provisionally Accepted

Chaoyi Wang¹ Zuobin Ying^1* Zijie Pan^1*

• ¹City University of Macau, Macao, SAR China

Machine unlearning, essential for data privacy and regulatory compliance, involves selectively removing specific information from a machine learning model. This study introduces an innovative method for machine unlearning in Spiking Neuron Models (SNMs), which more accurately replicate biological neural network behaviors. We adopt a hybrid approach that integrates selective synaptic retraining, synaptic pruning, and adaptive neuron thresholding. This method effectively eliminates targeted information while maintaining the neural network's overall integrity and performance. We conducted extensive experiments on various computer vision datasets to evaluate the impact of machine unlearning on critical performance metrics such as accuracy, precision, recall, and ROC AUC. Our findings confirm the practicality and efficiency of our approach, underscoring its applicability in real-world AI systems. This research contributes significantly to understanding machine unlearning in intricate neural architectures and paves the way for further developments in creating flexible and ethical AI models.