Research map / Label-efficient and robust learning / Recognition and long-tail
Explanation by Progressive Exaggeration
Sumedha Singla, Brian Pollack, Junxiang Chen, Kayhan Batmanghelich
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 |
|---|---|---|---|
| Use progressive exaggeration instead of xGEM counterfactual generation | CelebA: Smiling · CelebA: Smiling, overall | FID: 106.3 -> 35.8 | Improved |
| Use progressive exaggeration instead of xGEM counterfactual generation | CelebA: Young · CelebA: Young, overall | FID: 117.9 -> 53.4 | Improved |
| Use progressive exaggeration instead of xGEM counterfactual generation | Xray: Cardiomegaly · Xray: Cardiomegaly, overall | FID: 326.3 -> 58.1 | Improved |
| Use progressive exaggeration instead of xGEM | CelebA face explanations · CelebA: Smiling | Face verification accuracy: 0.0 -> 85.3 | Improved |
| Use progressive exaggeration instead of xGEM | CelebA face explanations · CelebA: Young | Face verification accuracy: 0.0 -> 72.2 | Improved |
| Add the adversarial data-consistency loss at lambda_cGAN=1 rather than removing it or using larger weights (ablation) | Explanation by Progressive Exaggeration · CelebA: Young vs Old | Overall FID: 67.2; 53.4; 63.0; 44.26 | Mixed |
| Add the adversarial data-consistency loss at lambda_cGAN=1 (ablation) | Explanation by Progressive Exaggeration · CelebA: Young vs Old | Latent Space Closeness: 96.1 -> 81.6 | Hurt |
| Add classifier-compatibility loss at lambda_f=1 rather than removing it or using larger weights (ablation) | Explanation by Progressive Exaggeration · CelebA: Young vs Old | Overall FID: 44.9; 53.4; 62.4; 71.2 | Mixed |
| Add classifier-compatibility loss at lambda_f=1 (ablation) | Explanation by Progressive Exaggeration · CelebA: Young vs Old | Face verification accuracy: 99.4 -> 72.2 | Hurt |
| Add reconstruction and cycle-consistency loss at lambda_rec=100 rather than removing it or using lower weights (ablation) | Explanation by Progressive Exaggeration · CelebA: Young vs Old | Overall FID: 72.2; 61.6; 65.7; 53.4 | Mixed |
| Add reconstruction and cycle-consistency loss at lambda_rec=100 (ablation) | Explanation by Progressive Exaggeration · CelebA: Young vs Old | Face verification accuracy: 0.0 -> 72.2 | Hurt |
Closest papers by meaning
- Interpretable Explanations of Black Boxes by Meaningful Perturbation (ICCV 2017 · similarity 0.87)
- Explanation-based Data Augmentation for Image Classification (NeurIPS 2021 · similarity 0.86)
- Explaining in Diffusion: Explaining a Classifier with Diffusion Semantics (CVPR 2025 · similarity 0.84)
- Selective Explanations (NeurIPS 2024 · similarity 0.83)
- I-CEE: Tailoring Explanations of Image Classification Models to User Expertise (AAAI 2024 · similarity 0.83)
- On Gradient-like Explanation under a Black-box Setting: When Black-box Explanations Become as Good as White-box (ICML 2024 · similarity 0.83)
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.
