AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Improving Low-Precision Network Quantization via Bin Regularization

Tiantian Han, Dong Li, Ji Liu, Lu Tian, Yi Shan

ICCV 2021 · Architectures and efficiency › Quantization · approach: Mixed-precision quantization · 35 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 bin regularizationLSQ, 2-bit · MobileNetV2 ImageNettop-1 accuracy: 46.7 -> 50.6Improved
add bin regularizationLSQ, 2-bit · MobileNetV3-Small ImageNettop-1 accuracy: 31.4 -> 36.3Improved
add bin regularizationLSQ, 3-bit · MobileNetV3-Small ImageNettop-1 accuracy: 52.0 -> 56.0Improved
add bin regularizationLSQ, 2-bit · ResNet18 ImageNettop-1 accuracy: 66.5 -> 67.2Improved
add bin regularizationLSQ, 4-bit · ResNet18 ImageNettop-1 accuracy: 70.5 -> 70.8Improved
add bin regularizationLSQ, 4-bit · MobileNetV2 ImageNettop-1 accuracy: 69.5 -> 70.4Improved
use KURE regularization instead of BR (ablation)LSQ, 2-bit · MobileNetV2 ImageNettop-1 accuracy: 46.7 -> 37.0Hurt
use L2 regularization instead of BR (ablation)LSQ, 2-bit · MobileNetV2 ImageNettop-1 accuracy: 46.7 -> 42.1Hurt
use L2 regularization instead of BR (ablation)LSQ, 3-bit · MobileNetV2 ImageNettop-1 accuracy: 65.3 -> 66.5Improved
add bin regularization from the start jointly (S1) instead of after 30 epochs of step-size learning (S2) (ablation)LSQ + BR, 2-bit · MobileNetV2 ImageNettop-1 accuracy: 50.6 -> 44.3Hurt
alternate step-size updates and bin regularization (S3) instead of S2 (ablation)LSQ + BR, 2-bit · MobileNetV2 ImageNettop-1 accuracy: 50.6 -> 43.6Hurt
vary loss weight lambda from 4 to 0.05 (ablation)LSQ + BR, 3-bit · MobileNetV3-Small ImageNettop-1 accuracy: 55.2 -> 54.7Mixed

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