AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Differentiable Dynamic Quantization with Mixed Precision and Adaptive Resolution

Zhaoyang Zhang, Wenqi Shao, Jinwei Gu, Xiaogang Wang, Ping Luo

ICML 2021 · 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 DDQ for 4-bit weights and 8-bit activations instead of uniform quantizationMobileNetV2 · ImageNetTop-1 accuracy: 67.1 -> 71.2Improved
Use DDQ for 4-bit weights and 8-bit activations instead of power-of-two quantizationMobileNetV2 · ImageNetTop-1 accuracy: 69.2 -> 71.2Improved
Add gradient correction to fixed-precision DDQ (ablation)MobileNetV2 · ImageNet, 4/8-bitTop-1 accuracy: 71.2 -> 71.6Improved
Add gradient correction to fixed-precision DDQ (ablation)ResNet18 · ImageNet, 4/8-bitTop-1 accuracy: 70.8 -> 70.9Improved
Use mixed rather than fixed bitwidth under the 4-bit target (ablation)DDQ on MobileNetV2 · ImageNetTop-1 accuracy: 70.7 -> 71.5Improved
Use mixed rather than fixed bitwidth under the 4-bit target (ablation)DDQ on ResNet18 · ImageNetTop-1 accuracy: 70.7 -> 71.0Improved
Learn adaptive quantization levels instead of using uniform levels (ablation)DDQ on MobileNetV2 · ImageNet, 4/4-bitTop-1 accuracy: 65.2 -> 70.7Improved
Learn adaptive quantization levels instead of using power-of-two levels (ablation)DDQ on MobileNetV2 · ImageNet, 4/4-bitTop-1 accuracy: 68.6 -> 70.7Improved
Use DDQ with 2/2-bit quantization (ablation)MobileNetV2 · ImageNetTop-1 accuracy: 71.9 -> 55.4Hurt
Use DDQ mixed precision on mobile DSP instead of full precisionMobileNetV2 · Mobile DSPLatency: 7.8 -> 5.1Improved
Use DDQ mixed precision instead of uniform quantizationMobileNetV2 · Mobile DSPTop-1 accuracy: 67.1 -> 71.9Improved
Use mixed precision with uniform quantizer constraints instead of original DDQ (ablation)MobileNetV2 · Mobile DSPTop-1 accuracy: 71.9 -> 70.8Hurt

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