Backward Reduction of CNN Models with Information Flow Analysis
share
This paper proposes backward reduction, an algorithm that explores the compact CNN design from the information flow perspective. This algorithm can remove substantial non-zero weighting parameters (redundant neural channels) by considering the network dynamic behavior, which the traditional model compaction techniques cannot achieve, to reduce the size of a model. With the aid of our proposed algorithm, we achieve significant model reduction results of ResNet-34 in ImageNet scale (32.3 state-of-the-art result (10.8 SqueezeNet and MobileNet, we still achieve additional 10.81 reduction, respectively, with negligible performance degradation.
READ FULL TEXT
