Abstract
AbstractDeep Convolutional Neural Networks (CNNs) have high memory footprint and computing power requirements, making their deployment in embedded devices difficult. Network pruning has received attention in reducing those requirements of CNNs. Among the pruning methods, Stripe-Wise Pruning (SWP) achieved a further network compression than conventional filter pruning methods and can obtain optimal kernel shapes of filters. However, the model pruned by SWP has filter redundancy because some filters have the same kernel shape. In this paper, we propose the Filter Shape Pruning (FSP) method, which prunes the networks using the kernel shape while maintaining the receptive fields. To obtain an architecture that satisfies the target FLOPs with the FSP method, we propose the Adaptive Architecture Search (AAS) framework. The AAS framework adaptively searches for the architecture that satisfies the target FLOPs with the layer-wise threshold. The layer-wise threshold is calculated at each iteration using the metric that reflects the filter’s influence on accuracy and FLOPs together. Comprehensive experimental results demonstrate that the FSP can achieve a higher compression ratio with an acceptable reduction in accuracy. KeywordsDeep learning optimizationStructured pruningConvolution neural networks
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have