AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Toward Faster and Simpler Matrix Normalization via Rank-1 Update

Tan Yu, Yunfeng Cai, Ping Li

ECCV 2020 · Architectures and efficiency › RNNs and normalization · approach: Differentiable optimization layers · 18 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 RUN normalization (eta=0.6 vs eta=0.0) (ablation)bilinear pooling · CUBaccuracy (%): 84.1 -> 86.3Improved
add RUN normalization (eta=0.6 vs eta=0.0) (ablation)bilinear pooling · DTDaccuracy (%): 65.6 -> 68.4Improved
add RUN normalization (eta=0.6 vs eta=0.0) (ablation)bilinear pooling · MITaccuracy (%): 79.8 -> 80.8Improved
add RUN normalization (eta=0.6 vs eta=0.0) (ablation)bilinear pooling · Aircraftaccuracy (%): 88.9 -> 89.8No clear effect
set eta=2.0 instead of 0.6 (ablation)RUN · CUBaccuracy (%): 86.3 -> 83.9Hurt
vary eta within [0.4, 1.5] (ablation)RUN · CUBaccuracy (%): 86.0 -> 86.4No clear effect
use K=2 power iterations instead of 5 (ablation)RUN · Aircraftaccuracy (%)No clear effect
use Tensor Sketch instead of Random Maclaurin for CBP (ablation)RUN + CBP · CUBaccuracy (%)No clear effect
increase compact feature dimension from 1K to 10K (ablation)RUN + CBP · CUBaccuracy (%)Improved
replace NS iteration with RUN for normalizationbilinear pooling · GPU time per batchGPU time (ms): 833ms -> 2.5msImproved
replace SVD in MoNet-2 with RUNcompact bilinear pooling · GPU time per batchGPU time (ms): 13850ms -> 2.5msImproved
replace SVD in MoNet-2 with RUNcompact bilinear pooling · Aircraftaccuracy (%): 86.7 -> 91.0Improved

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