Extracting coarse-grained classifiers from large Convolutional Neural Networks

Katarzyna Filus, Joanna Domanska

Abstract: Highlights•A novel method for reusing fine-grained models to create coarse-grained models.•Reusing the existing classifiers with relatively good accuracy with no training.•Three different strategies based on concepts’ similarity and hierarchy are proposed.•The method can be used to create models with labels at multiple hierarchy levels.