Pruning CNN filters via quantifying the importance of deep visual representations

Ali Alqahtani, Xianghua Xie, Mark W. Jones, Ehab Essa

Abstract: Highlights•A novel framework to prune the unimportant filters to compress and accelerate CNNs.•A pioneering attempt to apply the interpretability for effective CNNs compression.•Measuring the expressive feature across different filters•A majority voting method to assign a voting score for filters importance measurement.•A convolution kernels estimation method to minimize the damage of the pruning procedure.