Frequency-Domain Dynamic Pruning for Convolutional Neural Networks

Authors: Zhenhua Liu, Jizheng Xu, Xiulian Peng, Ruiqin Xiong

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that the proposed scheme can outperform previous spatial-domain counterparts by a large margin. Specifically, it can achieve a compression ratio of 8.4 and a theoretical inference speed-up of 9.2 for Res Net-110, while the accuracy is even better than the reference model on CIFAR-10.
Researcher Affiliation Collaboration Zhenhua Liu1, Jizheng Xu2, Xiulian Peng2, Ruiqin Xiong1 1Institute of Digital Media, School of Electronic Engineering and Computer Science, Peking University 2Microsoft Research Asia
Pseudocode No The paper states: 'The process of the proposed algorithm is detailed in the supplementary material.' However, no pseudocode or algorithm blocks are provided within the main body of the paper.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository for the methodology described.
Open Datasets Yes In this section, we conduct comprehensive experiments on three benchmark datasets: MNIST, Image Net and CIFAR-10.
Dataset Splits Yes We conduct comprehensive experiments on three benchmark datasets: MNIST, Image Net and CIFAR-10... Alex Net on Image Net: ILSVRC-2012 dateset, which has 1.2 million training images and 50K validation images... Res Net on CIFAR-10: CIFAR-10 is also a classification benchmark dataset containing a training set of 50K images and a test set of 10K images.
Hardware Specification No The paper does not provide specific details about the hardware used for the experiments, such as GPU or CPU models. It only mentions that 'The training processes are all performed with the Caffe framework'.
Software Dependencies No The paper mentions: 'The training processes are all performed with the Caffe framework [24].' However, it does not specify any version numbers for Caffe or any other software dependencies.
Experiment Setup Yes The momentum and weight decay are set to 0.9 and 0.0001 in all experiments... The learning rate is set to 0.1 initially and reduced by 10 times for every 4K iterations during training. We use 'xavier' initialization method and train a reference model whose top-1 accuracy is 99.07% with 10K iterations... batch size is set to 64... Alex Net: We use a learning rate of 0.01 and reduced it by 10 times for every 100K iterations. The batch size is set to 32 and we use 'gaussian' initialization method... Res Net: The learning rate is 0.1 and reduced by 10 times for every 40K iterations. The 'msra' initialization method is adopted in this experiment. The batch size is set to 100 and the maximal number of iterations is set to 150K.