AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Hierarchical Channel-spatial Encoding for Communication-efficient Collaborative Learning

Qihua Zhou, Song Guo, LIU Yi, Jie Zhang, Jiewei Zhang, Tao GUO, Zhenda XU, XUN LIU, Zhihao Qu

NeurIPS 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
replace uniform quantization with stripe-wise group quantization (SGQ)feature compression · mini-ImageNet ResNet34, 4-bitaccuracy (%): 36.89 -> 74.37Improved
replace product quantization with SGQfeature compression · mini-ImageNet ResNet34, 4-bitaccuracy (%): 69.61 -> 74.37Improved
replace PQ with SGQfeature compression · mini-ImageNet ShuffleNet-x1.0, 4-bitaccuracy (%): 70.16 -> 74.86Improved
replace PQ with SGQfeature compression · CIFAR-10 MobileNet-V1, 4-bitaccuracy (%): 92.67 -> 93.45Improved
replace PQ with SGQfeature compression · Fashion-MNIST VGG11, 4-bitaccuracy (%): 95.94 -> 96.57Improved
replace CLIO top-slice dropping with SGQ at same compressionfeature compression · Fashion-MNIST VGG11, 4-bitaccuracy (%): 21.02 -> 96.57Improved
SGQ 4-bit compression versus FP32 trainingSGQ · mini-ImageNet ResNet34accuracy (%): 80.31 -> 74.37Hurt
SGQ 4-bit compression versus FP32 trainingSGQ · Fashion-MNIST VGG11accuracy (%): 97.55 -> 96.57Mixed
lower quantization bits from 8 to 4SGQ · system overheadtotal overhead proportion (%): 43.57 -> 9.89Improved
lower quantization bits from 4 to 2SGQ · system overheadtotal overhead proportion (%): 9.89 -> 8.14Improved
use pixel sampling via average pooling in clusteringSGQ module · system overhead, 4 bitsSGQ module overhead (%): 31.93 -> 5.51Improved
use fewer quantization bits (4,5,6) (ablation)SGQ · accuracy across benchmarksaverage accuracyMixed

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