AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Allowing Oscillation Quantization: Overcoming Solution Space Limitation in Low Bit-Width Quantization

Weiying Xie, Zihan Meng, Jitao Ma, Wenjin Guo, Haowei Li, Haonan Qin, Leyuan Fang, Yunsong Li

ICCV 2025 · Architectures and efficiency › Quantization · approach: Mixed-precision quantization · 1 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
add early-stage oscillation encouragement and decoupled thresholds/levels (expand solution space) (ablation)LSQ, ResNet-50 W2/A2 · ImageNet valtop-1 accuracy (%): 73.7 -> 75.2Improved
add delayed weight oscillation suppression (ablation)LSQ, ResNet-50 W2/A2 · ImageNet valtop-1 accuracy (%): 73.7 -> 75.6Improved
combine expanded solution space and delayed suppression (ablation)LSQ, ResNet-50 W2/A2 · ImageNet valtop-1 accuracy (%): 73.7 -> 76.3Improved
allow oscillation early instead of suppressing it throughout training (ablation)LSQ toy example · toy 2-bit problemminimum loss: 26.2 -> 5.6Improved
suppress oscillation only in final stage versus throughout (ablation)LSQ toy example · toy 2-bit problemminimum loss: 26.2 -> 16.7Improved
add AOQ to LSQ on 2-bit ResNet-18LSQ · ImageNet val ResNet-18 W2/A2top-1 accuracy (%): 67.6 -> 69.7Improved
add AOQ to N2UQ on 4-bit ResNet-18N2UQ · ImageNet val ResNet-18 W4/A4top-1 accuracy (%): 72.9 -> 72.7Hurt
add AOQ to N2UQ on ResNet-34/50 at 3/4 bitsN2UQ · ImageNet val ResNet-34 W4/A4top-1 accuracy (%): 76.0 -> 76.0No clear effect
use AOQ on DeiT-T at 3-bit versus OFQLSQ+AOQ · ImageNet val DeiT-T W3/A3top-1 accuracy (%): 72.7 -> 72.3Hurt
use AOQ instead of OFQ on 2-bit DeiT-TLSQ+AOQ · ImageNet val DeiT-T W2/A2top-1 accuracy (%): 64.3 -> 66.5Improved
use AOQ instead of oscillation suppression throughout training on MobileNet-V2LSQ+AOQ · ImageNet val MobileNet-V2 W3/A3top-1 accuracy (%): 67.8 -> 69.1Improved

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