Feature Statistics Guided Efficient Filter Pruning

Authors: Hang Li, Chen Ma, Wei Xu, Xue Liu

IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive empirical experiments with various CNN architectures on publicly available datasets. The experimental results demonstrate that our model obtains up to 91.6% parameter decrease and 83.7% FLOPs reduction with almost no accuracy loss.
Researcher Affiliation Academia 1School of Computer Science, Mc Gill University 2Institute for Interdisciplinary Information Sciences, Tsinghua University
Pseudocode Yes Algorithm 1 Our proposed filter pruning scheme; Algorithm 2 Similarity-aware feature map selection (SFS)
Open Source Code No The paper does not contain any explicit statement about releasing source code or provide a link to a code repository.
Open Datasets Yes CIFAR10 and CIFAR100 [Krizhevsky et al., 2009] are two widely used datasets with 32 × 32 colour natural images. They both contain 50, 000 training images and 10, 000 test images with 10 and 100 classes respectively. ... ILSVRC-2012 is a large-scale dataset with 1.2 million training images and 50, 000 validation images of 1000 classes.
Dataset Splits Yes For ILSVRC-2012, we use the pre-trained Res Net-50 released by Pytorch. We train Mobile Net for 60 epochs with a weight decay of 0.0015. ... ILSVRC-2012 is a large-scale dataset with 1.2 million training images and 50, 000 validation images of 1000 classes.
Hardware Specification No The paper does not provide specific details about the hardware used, such as GPU/CPU models or memory specifications.
Software Dependencies No The paper mentions using 'Pytorch' but does not specify its version number or any other software dependencies with version details.
Experiment Setup Yes For CIFAR, we set the mini-batch size to 64, epochs to 160 with a weight decay of 0.0015 and Nesterov momentum [Sutskever et al., 2013] of 0.9. For ILSVRC-2012, we use the pre-trained Res Net-50 released by Pytorch. We train Mobile Net for 60 epochs with a weight decay of 0.0015.