Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Interpretable Model with Transformation Equivariant Interpretation
Yipei Wang, Xiaoqian Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add transformation-equivariance regularization to learned prototypes | SITE · CIFAR-10 validation | interpretation self-consistency | Improved |
| Use SITE interpretations instead of post-hoc Grad-CAM interpretations | CIFAR-10 validation (transformed) · CIFAR-10 validation | self-consistency score: 0.8817 -> 0.8860 | Improved |
| Use SITE interpretations instead of post-hoc Guided BP interpretations | CIFAR-10 validation (transformed) · CIFAR-10 validation | self-consistency score: 0.7830 -> 0.8860 | Improved |
| Use SITE interpretations instead of post-hoc Excitation BP interpretations | CIFAR-10 validation (transformed) · CIFAR-10 validation | self-consistency score: 0.1159 -> 0.8860 | Improved |
| Use SITE interpretations instead of post-hoc Gradient interpretations | CIFAR-10 validation (transformed) · CIFAR-10 validation | self-consistency score: 0.7174 -> 0.8860 | Improved |
| Use SITE interpretations instead of post-hoc Linear Approx. interpretations | CIFAR-10 validation (transformed) · CIFAR-10 validation | self-consistency score: 0.8485 -> 0.8860 | Improved |
| Use SITE interpretations instead of post-hoc DeConvNet interpretations | CIFAR-10 validation (transformed) · CIFAR-10 validation | self-consistency score: 0.8591 -> 0.8860 | Improved |
| Use SITE rather than a black-box CNN with the same backbone | MNIST validation · MNIST | accuracy: 0.981 -> 0.988 | Improved |
| Use SITE rather than a black-box ResNet with the same backbone | CIFAR-10 validation · CIFAR-10 | accuracy: 0.862 -> 0.892 | Improved |
| Use SITE rather than a black-box ResNet-18 | Food-101 validation · Food-101 | accuracy: 64.04% -> 63.91% | Hurt |
| Use SITE rather than Linear Approx. for explanations | MNIST annotated transformed images · MNIST pointing game | hit ratio: 0.6540 -> 0.9993 | Improved |
| Use SITE rather than Guided BP for explanations | MNIST annotated untransformed images · MNIST pointing game | hit ratio: 0.8825 -> 0.9996 | Improved |
Closest papers by meaning
- Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance (NeurIPS 2023 · similarity 0.88)
- Interpreting Equivariant Representations (ICML 2024 · similarity 0.85)
- Self-supervised Transformation Learning for Equivariant Representations (NeurIPS 2024 · similarity 0.85)
- Understanding the Role of Equivariance in Self-supervised Learning (NeurIPS 2024 · similarity 0.85)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 2024 · similarity 0.84)
- Contrastive-Equivariant Self-Supervised Learning Improves Alignment with Primate Visual Area IT (NeurIPS 2024 · similarity 0.84)
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.
