AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Self-Interpretable Model with Transformation Equivariant Interpretation

Yipei Wang, Xiaoqian Wang

NeurIPS 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Self-supervision in supervised pipelines

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 transformation-equivariance regularization to learned prototypesSITE · CIFAR-10 validationinterpretation self-consistencyImproved
Use SITE interpretations instead of post-hoc Grad-CAM interpretationsCIFAR-10 validation (transformed) · CIFAR-10 validationself-consistency score: 0.8817 -> 0.8860Improved
Use SITE interpretations instead of post-hoc Guided BP interpretationsCIFAR-10 validation (transformed) · CIFAR-10 validationself-consistency score: 0.7830 -> 0.8860Improved
Use SITE interpretations instead of post-hoc Excitation BP interpretationsCIFAR-10 validation (transformed) · CIFAR-10 validationself-consistency score: 0.1159 -> 0.8860Improved
Use SITE interpretations instead of post-hoc Gradient interpretationsCIFAR-10 validation (transformed) · CIFAR-10 validationself-consistency score: 0.7174 -> 0.8860Improved
Use SITE interpretations instead of post-hoc Linear Approx. interpretationsCIFAR-10 validation (transformed) · CIFAR-10 validationself-consistency score: 0.8485 -> 0.8860Improved
Use SITE interpretations instead of post-hoc DeConvNet interpretationsCIFAR-10 validation (transformed) · CIFAR-10 validationself-consistency score: 0.8591 -> 0.8860Improved
Use SITE rather than a black-box CNN with the same backboneMNIST validation · MNISTaccuracy: 0.981 -> 0.988Improved
Use SITE rather than a black-box ResNet with the same backboneCIFAR-10 validation · CIFAR-10accuracy: 0.862 -> 0.892Improved
Use SITE rather than a black-box ResNet-18Food-101 validation · Food-101accuracy: 64.04% -> 63.91%Hurt
Use SITE rather than Linear Approx. for explanationsMNIST annotated transformed images · MNIST pointing gamehit ratio: 0.6540 -> 0.9993Improved
Use SITE rather than Guided BP for explanationsMNIST annotated untransformed images · MNIST pointing gamehit ratio: 0.8825 -> 0.9996Improved

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