AI Paper Atlas

Research map / Architectures and efficiency / Quantization

SDQ: Stochastic Differentiable Quantization with Mixed Precision

Xijie Huang, Zhiqiang Shen, Shichao Li, Zechun Liu, Hu Xianghong, Jeffry Wicaksana, Eric Xing, Kwang-Ting Cheng

ICML 2022 · Architectures and efficiency › Quantization · approach: Mixed-precision quantization · 7 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
use stochastic differentiable bitwidth selection instead of the strongest listed fixed-precision or MPQ baselinesResNet20 · CIFAR-10Top-1 accuracy (%): 91.9 -> 92.1Improved
use SDQ mixed precision instead of DDQResNet18 · ImageNet-1KTop-1 accuracy (%): 71.2 -> 71.7Improved
use SDQ mixed precision instead of DDQMobileNetV2 · ImageNet-1KTop-1 accuracy (%): 71.8 -> 72.0Improved
use SDQ strategy generation rather than HMQ strategy generation (ablation)MobileNetV2 · ImageNet-1KTop-1 accuracy (%): 71.7 -> 72.0Improved
use SDQ strategy generation rather than FracBits strategy generation (ablation)MobileNetV2 · ImageNet-1KTop-1 accuracy (%): 72.0 -> 72.0No clear effect
add EBR with lambda 0.1 (ablation)mixed-precision ResNet18 · ImageNet-1KTop-1 accuracy (%): 67.6 -> 69.1Improved
replace EBR with Weight Norm (ablation)mixed-precision ResNet18 · ImageNet-1KTop-1 accuracy (%): 69.1 -> 66.6Hurt
vary EBR weight from 0.1 to 0.01 (ablation)mixed-precision ResNet18 · ImageNet-1KTop-1 accuracy (%): 69.1 -> 68.6Hurt
vary EBR weight from 0.1 to 1 (ablation)mixed-precision ResNet18 · ImageNet-1KTop-1 accuracy (%): 69.1 -> 68.9Hurt
add knowledge distillation from a ResNet101 teacher (ablation)mixed-precision ResNet18 · ImageNet-1KTop-1 accuracy (%): 70.5 -> 71.7Improved
use a stronger ResNet101 teacher rather than ResNet34 (ablation)mixed-precision ResNet18 · ImageNet-1KTop-1 accuracy (%): 70.7 -> 71.7Improved
use block-wise rather than layer-wise quantization (ablation)ResNet18 · ImageNet-1KTop-1 accuracy (%): 71.7 -> 71.2Hurt

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