Accepted to Efficient Large Vision Models @ CVPR 2025
Authors: Nick Cui, Trevor Xing-Xie, Arushi Gupta, Peter Choi
Inspired by the critical brain hypothesis in neuroscience, this paper explores and leverages centrality metrics to facilitate efficient pruning in deep neural networks. We develop a novel pruning methodology that uses neuroscience-informed centrality measures to identify and retain critical connections while removing redundant pathways. Our approach achieves competitive compression rates while maintaining model accuracy, demonstrating that biological principles can inform more efficient artificial neural network design. [arXiv link TBA]
