Feature Map Analysis-Based Dynamic CNN Pruning and the Acceleration on FPGAs

Deep-learning-based applications bring impressive results to graph machine learning and are widely used in fields such as autonomous driving language translations. Nevertheless, the tremendous capacity of convolutional neural networks makes it difficult for them be implemented on resource-constrained devices. Channel pruning provides a promising solution compress by removing redundant calculation. Existing methods measure importance each filter discard less important ones until reaching fixed compression target. However, static approach limits effect. Thus, we propose dynamic channel-pruning method that dynamically identifies removes filters based redundancy analysis its feature maps. Experimental show 77.10% floating-point operations per second (FLOPs) 91.72% parameters reduced VGG16BN with only 0.54% accuracy drop. Furthermore, compressed models were field-programmable gate array (FPGA) significant speed-up was observed.