AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Computation Error Analysis of Block Floating Point Arithmetic Oriented Convolution Neural Network Accelerator Design

Zhourui Song, Zhenyu Liu, Dongsheng Wang

AAAI 2018 · Architectures and efficiency › Quantization · approach: Low-precision training formats · 39 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 BFP with 8-bit weight and input mantissasVGG-16 · VGG-16accuracy drop: 0.0002No clear effect
Use BFP with 8-bit weight and input mantissasResNet-18 · ResNet-18accuracy drop: 0.0026Hurt
Use BFP with 8-bit weight and input mantissasResNet-50 · ResNet-50accuracy drop: 0.0008No clear effect
Increase input mantissa length from 6 to 8 bits at fixed 8-bit weight mantissa (ablation)VGG-16 · VGG-16accuracy drop: 0.1772 -> 0.0002Improved
Increase weight mantissa length from 6 to 8 bits at fixed 8-bit input mantissa (ablation)VGG-16 · VGG-16accuracy drop: 0.1246 -> 0.0002Improved
Increase input mantissa length from 6 to 8 bits at fixed 8-bit weight mantissa (ablation)ResNet-50 · ResNet-50accuracy drop: 0.0664 -> 0.0008Improved
Increase input mantissa length from 7 to 8 bits at fixed 7-bit weight mantissa (ablation)CIFAR-10 · CIFAR-10accuracy drop: 0.0042 -> 0.0028Improved
Increase input mantissa length from 6 to 8 bits at fixed 7-bit weight mantissa (ablation)CIFAR-10 · CIFAR-10accuracy drop: 0.0042 -> 0.0014Improved
Choose the second tested BFP partition rather than the first (ablation)VGG-16 · VGG-16, ILSVRC12top-1 accuracy: 0.6672 -> 0.6832Improved
Use BFP partition choice 2 rather than floating-point representationVGG-16 · VGG-16, ILSVRC12top-1 accuracy: 0.6808 -> 0.6832No clear effect
Round rather than truncate shifted-out mantissa bits (ablation)BFP CNN arithmetic · CNN convolutionclassification accuracyImproved
Use an 8-bit rather than shorter mantissa for deep CNNs (ablation)VGG-16, GoogLeNet, ResNet-18, ResNet-50 · ILSVRC12accuracy drop: less than 0.3%Improved

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