AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Towards Accurate Post-training Network Quantization via Bit-Split and Stitching

Peisong Wang, Qiang Chen, Xiangyu He, Jian Cheng

ICML 2020 · Architectures and efficiency › Quantization · approach: Post-training weight quantization

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
Bit-split weight quantization (vs TF-Lite) at 3-bit, weights onlyResNet-18/50/101, VGG-16-BN weights · ImageNet classificationtop-5 accuracy: ResNet-18 87.45 vs 2.68 for TF-LiteImproved
Bit-split weight quantization vs TF-Lite at 4-bit and higher, weights onlyweight quantization · ImageNet classificationtop-1 accuracy: gap grows as bit-width decreasesImproved
Per-channel activation quantization vs per-layer (ResNet-18, 8-bit weights) (ablation)activation quantization · ImageNet ResNet-18top-1 accuracy: 3-bit: 65.91 vs 64.77Improved
ECAQ (error compensated per-layer activation quantization) vs direct per-layer quantization (ablation)activation quantization · ImageNet ResNet-18, 3-bit and 4-bit activationstop-1 accuracy: +1.44% at 3-bit, +0.45% at 4-bitImproved
ECAQ vs direct per-layer quantization at 8-bit to 5-bit activations (ablation)activation quantization · ImageNet ResNet-18top-1 accuracy: differences less than 0.1% at higher bit widthsNo clear effect
Bit-split 8-bit activation quantization added to 8-bit-weight... (A8 per-layer) on top of Bit-split weights (ablation)ResNet-18 activations · ImageNet classification, weights 3-8 bittop-1 accuracy: differences under 0.1%No clear effect
Bit-split + ECAQ joint quantization of weights and activations at 4-bit vs TF-LiteResNet-18, ResNet-50, VGG-16-BN · ImageNet classification W4A4top-5 accuracy: ResNet-18 87.76 vs 36.33Improved
Bit-split with 4-bit weights and 8-bit activations (per-layer) vs ACIQ-Mix (mixed precision, per-channel, dynamic)ResNet-18 A8W4 · ImageNet classificationtop-1 accuracy: 69.1 vs 68.3 (Bit-split higher on ResNet-18); ResNet-50 75.6 vs 75.3, but ResNet-101 76.9 tie and VGG-16-BN 73.0 vs 72.4Mixed
Per-channel activation quantization added to Bit-split (W4A4) (ablation)Bit-split W4A4 · ImageNet classificationtop-1 accuracy: ResNet-18 68.1 vs 67.6 (per-layer)Improved
Bit-split 4-bit weights with 8-bit activations applied to detection/segmentationRetinaNet and Mask R-CNN (ResNet-50) · COCO minivalAP0.5:0.95: about 0.8%~1.2% mAP degradationNo clear effect

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