AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

GRIM: A General, Real-Time Deep Learning Inference Framework for Mobile Devices Based on Fine-Grained Structured Weight Sparsity

Wei Niu, Zhengang Li, Xiaolong Ma, Peiyan Dong, Gang Zhou, Xuehai Qian, Xue Lin, Yanzhi Wang, Bin Ren

TPAMI 2021 · Architectures and efficiency › Pruning and efficient tuning · approach: Sparse training from scratch · 23 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
replace pattern-based pruning (PatDNN) with BCR pruning at higher rateVGG-16 · CIFAR-10accuracy: 93.7 -> 93.8 (19.8x -> 35.7x)Improved
replace pattern-based pruning (PatDNN) with BCR pruningResNet-18 · CIFAR-10 27.0xaccuracy: 92.1 -> 93.9Improved
replace filter pruning (DCP) with BCR pruningResNet-18 · CIFAR-10 27.0xaccuracy: 68.2 -> 93.9Improved
raise BCR pruning rate from 35.7x to 71.3xGRIM VGG-16 · CIFAR-10accuracy: 93.8 -> 93.1Hurt
raise BCR pruning rate from 4.0x to 8.0xGRIM ResNet-18 · ImageNettop-1 accuracy: 69.6 -> 67.2Hurt
raise BCR pruning rate from 19.5x to 245.5xGRIM GRU · TIMITphone error rate: 18.8 -> 24.2Hurt
use BCR pruning instead of block-circulant (C-LSTM) at 16xGRU · TIMITphone error rate: 25.48 -> 18.8Improved
add matrix reorder to BCR-pruned code (CPU) (ablation)GRIM · VGG-16 ImageNet layersspeedup: 1.21x to 1.88x speedupImproved
add matrix reorder to BCR-pruned code (GPU) (ablation)GRIM · VGG-16 ImageNet layersspeedup: 1.30x to 2.88x speedupImproved
add register load redundancy elimination (CPU) (ablation)GRIM · VGG-16 ImageNet layersspeedup: 1.11x to 3.51x extra speedupImproved
add register load redundancy elimination (GPU) (ablation)GRIM · VGG-16 ImageNet layersspeedup: 0.89x to 1.90x extra speedupMixed
add auto-tuning (CPU) (ablation)GRIM · VGG-16 ImageNet layersspeedup: 0.31x to 1.45x additional speedupMixed

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