Pruning Neural Networks Using Cooperative Game Theory

We introduce Game Theoretic Assisted Pruning (GTAP), a method that utlizes power indices from cooperative game theory to efficiently prune deep neural networks without compromising their predictive performance. GTAP identifies and removes less impactful neurons based on their contribution to the network's performance, streamlining the model's size and computational load. Our empirical evaluations show that GTAP outperforms traditional pruning techniques, achieving a better balance between model compactness and accuracy across multiple types of neural networks.