AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Generating Attribution Maps With Disentangled Masked Backpropagation

Adria Ruiz, Antonio Agudo, Francesc Moreno-Noguer

ICCV 2021 · Architectures and efficiency › RNNs and normalization · approach: Representation geometry and collapse · 5 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 disentanglement of positive and negative factors (optimize masks) versus vanilla gradient x input (ablation)DMBP · ImageNet (5K subset) ResNet50IM: .26 -> .50Improved
add disentanglement versus vanilla gradient x input (ablation)DMBP · VOC2012 val VGG16IM: .36 -> .62Improved
add negative-term minimization to positive-only loss (ablation)DMBP · VOC2012 val VGG16cIM: .06 -> .18Improved
add negative-term minimization to positive-only loss (ablation)DMBP · VOC2012 val VGG16IM: .62 -> .51Hurt
add negative-term minimization to positive-only loss (ablation)DMBP · VOC2012 val ResNet50IM: .58 -> .59Improved
add nuisance-term minimization (full loss) (ablation)DMBP · ImageNet (5K subset) ResNet50IM: .51 -> .57Improved
add nuisance-term minimization (full loss) (ablation)DMBP · ImageNet (5K subset) VGG16IM: .37 -> .39Improved
add nuisance-term minimization (full loss) (ablation)DMBP · VOC2012 val VGG16IM: .51 -> .57Hurt
DMBP versus GradCam on ImageNet VGG16DMBP · ImageNet val VGG16IM: 0.48 -> 0.41Hurt
DMBP versus GradCam average over all settingsDMBP · All settingsAvg of IM/cIM: 0.37 -> 0.42Improved
DMBP versus SmoothGrad on VOC VGG IMDMBP · VOC2012 val VGG16IM: 0.55 -> 0.57Improved
DMBP versus GBp in layer reinitialization sanity checkDMBP · VOC VGG16 reinit correlationrank correlation (lower better): 0.99 -> 0.02Improved

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