AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Finding the Task-Optimal Low-Bit Sub-Distribution in Deep Neural Networks

Runpei Dong, Zhanhong Tan, Mengdi Wu, Linfeng Zhang, Kaisheng Ma

ICML 2022 · Architectures and efficiency › Quantization · approach: Post-training weight quantization · 4 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 task-objective DGMS quantization instead of LQNetsResNet-20 and VGG-Small · CIFAR-10, 3-bitTop-1 accuracy: 92.00% -> 92.84%; 93.80% -> 94.46%Improved
Use DGMS quantization instead of TTQResNet-32 and ResNet-56 · CIFAR-10, 2-bitTop-1 accuracy: 92.37% -> 92.97%; 93.56% -> 93.72%Improved
Use DGMS quantization instead of LQNetsResNet-20 and VGG-Small · CIFAR-10, 2-bitTop-1 accuracy: 91.80% -> 92.13%; 93.80% -> 94.36%Improved
Use DGMS weight quantization instead of the listed baselinesResNet-18 · ImageNet, 4/32 and 4/4Top-1 accuracy: 70.00% -> 70.25%; 68.45% -> 68.95%Improved
Use DGMS weight quantization instead of HAQResNet-50 · ImageNet, 2/32Top-1 accuracy: 70.63% -> 72.88%Improved
Use DGMS weight quantization instead of HAQResNet-50 · ImageNet, 4/32Top-1 accuracy: 76.14% -> 76.28%Improved
Quantize detector weights with DGMSSSDLite-MBV2 and SSDLite-MBV3 · PASCAL VOC2007 testmAP: 68.6 -> 67.9; 67.2 -> 65.7Hurt
Deploy 4-bit models with Q-SIMD rather than primitive TVM schedulingDGMS quantized models · Inference on Qualcomm 888speedup: 7.46×; 4.34×; 1.74×; 1.66×; 1.72×Improved
Initialize gamma empirically rather than with standard deviation (ablation)DGMS · CIFAR-10, ResNet-20 2-bitTop-1 accuracy: 89.26% -> 92.13%Improved
Initialize gamma empirically rather than with standard deviation (ablation)DGMS · CIFAR-10, VGG-Small 2-bitTop-1 accuracy: 92.55% -> 94.41%Improved
Learn temperature instead of fixing it at initialization 0.010 (ablation)DGMS on VGG-Small 3-bit · CIFAR-10Top-1 accuracy: 94.56% -> 94.64%No clear effect
Increase the number of Gaussian mixture components (ablation)DGMS on ResNet-18 4-bit · ImageNetTop-1 accuracyImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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