Research map / Label-efficient and robust learning / Self-supervised and contrastive
AIM: Amending Inherent Interpretability via Self-Supervised Masking
Eyad Alshami, Shashank Agnihotri, Bernt Schiele, Margret Keuper
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 AIM[2, 25%] masking | ConvNeXt-tiny · Waterbirds-100% | worst-group accuracy: 39.6 -> 74.0 | Improved |
| add AIM[2, 35%] masking | ConvNeXt-tiny · Waterbirds-100% | overall accuracy: 71.2 -> 89.1 | Improved |
| add AIM[2, 25%] masking | ConvNeXt-tiny · Waterbirds-95% | worst-group accuracy: 81.6 -> 92.7 | Improved |
| add AIM[2, 25%] masking | ConvNeXt-tiny · TravelingBirds | accuracy: 59.5 -> 77.4 | Improved |
| add AIM[2, 25%] masking | ConvNeXt-tiny · TravelingBirds | EPG: 74.4 -> 85.0 | Improved |
| add AIM[2, 25%] masking | ConvNeXt-tiny · Hard-ImageNet | EPG: 36.6 -> 40.1 | Improved |
| add AIM[2, 25%] masking | ConvNeXt-tiny · ImageNet100 | accuracy: 89.2 -> 90.1 | No clear effect |
| add AIM[1, 25%] masking | ConvNeXt-tiny · CUB-200 | accuracy: 87.9 -> 88.6 | Improved |
| add AIM[1, 25%] masking | ConvNeXt-tiny · CUB-200 | EPG: 68.3 -> 76.804 | Improved |
| use top-down pathway AIM instead of bottom-up masking [1, 25%] (ablation) | ConvNeXt-tiny · CUB-200 | accuracy: 72.79 -> 88.82 | Improved |
| use top-down pathway AIM instead of bottom-up masking [2, 25%] (ablation) | ConvNeXt-tiny · CUB-200 | accuracy: 84.00 -> 88.677 | Improved |
| add AIM[1, 25%] on shifted-center crops | ConvNeXt-tiny · Shifted-center CUB-200 | accuracy: 76.98 -> 79.33 | Improved |
Closest papers by meaning
- Network Dissection: Quantifying Interpretability of Deep Visual Representations (CVPR 2017 · similarity 0.87)
- Q-SENN: Quantized Self-Explaining Neural Networks (AAAI 2024 · similarity 0.86)
- Measuring Per-Unit Interpretability at Scale Without Humans (NeurIPS 2024 · similarity 0.86)
- Exploring Target Representations for Masked Autoencoders (ICLR 2024 · similarity 0.86)
- A Multimodal Automated Interpretability Agent (ICML 2024 · similarity 0.86)
- Intrinsic User-Centric Interpretability through Global Mixture of Experts (ICLR 2025 · similarity 0.85)
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.
