AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Towards Fully Sparse Training: Information Restoration with Spatial Similarity

Weixiang Xu, Xiangyu He, Ke Cheng, Peisong Wang, Jian Cheng

AAAI 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Sparse training from scratch · 4 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add 2:4 sparsity on W, Wᵀ, and Xᵀ with Information RestorationFST with ResNet-18 · ImageNetTop-1 accuracy: 71.2 -> 71.0Hurt
Add 2:4 sparsity on W and WᵀFST with ResNet-18 · ImageNetTop-1 accuracy: 71.2 -> 71.3Improved
Add 2:4 sparsity on W, Wᵀ, and Xᵀ with Information RestorationFST with ResNet-50 · ImageNetTop-1 accuracy: 77.3 -> 77.0Hurt
Add 2:4 sparsity on W, Wᵀ, and Xᵀ with Information RestorationFST · PASCAL VOC, VGG16-SSDmAP: 75.8 -> 76.2Improved
Add fully sparse trainingR50 Faster-RCNN · PASCAL VOCmAP: 79.5 -> 79.4Hurt
Add fully sparse trainingVGG16-SSD · COCO2017mAP: 25.5 -> 25.5No clear effect
Add fully sparse trainingR50 Faster-RCNN · COCO2017mAP: 37.4 -> 37.2Hurt
Remove sparsity on W (ablation)FST with ResNet-18 · ImageNetTop-1 accuracy: 71.2 -> 71.2No clear effect
Add sparsity on Wᵀ (ablation)FST with ResNet-18 · ImageNetTop-1 accuracy: 71.2 -> 71.3Improved
Add fixed-position sparsity on Xᵀ without Information Restoration (ablation)FST with ResNet-18 · ImageNetTop-1 accuracy: 71.3 -> 69.4Hurt
Add Information Restoration (ablation)FST with ResNet-18 and sparse Xᵀ · ImageNetTop-1 accuracy: 69.4 -> 71.0Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy