Research map / Vision-language and multimodal / CLIP and zero-shot
Know When to Abstain: Optimal Selective Classification with Likelihood Ratios
Alvin Heng, Harold Soh
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 |
|---|---|---|---|
| estimate separate correct/wrong distributions (Delta-MDS vs MDS) | MDS · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 13.9 -> 7.81 | Improved |
| estimate separate correct/wrong distributions (Delta-KNN vs KNN) | KNN · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 13.1 -> 7.32 | Improved |
| average log-distance over top-k neighbors instead of k-th distance (ablation) | Delta-KNN · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 7.45 -> 7.32 | Improved |
| linearly combine Delta-KNN with RLog | Delta-KNN · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 7.32 -> 6.43 | Improved |
| linearly combine Delta-MDS with RLog | Delta-MDS · ImageNet variants avg (1K), EVA | AURC (x1e-2): 4.18 -> 3.86 | Improved |
| pair Delta-KNN with RLog instead of MSP (ablation) | Delta-KNN combination · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 7.18 -> 6.43 | Improved |
| combine two distance scores instead of distance plus logit (ablation) | Delta-KNN · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 6.43 -> 7.46 | Hurt |
| combine MSP with RLog (double-logit) (ablation) | RLog · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 7.39 -> 7.35 | No clear effect |
| use only 1% labeled samples for Delta-MDS statistics (ablation) | Delta-MDS-RLog · ImageNet-1K, DFN CLIP | AURC (x1e-2): 4.13 -> 10.5 | Hurt |
| use only 0.1% labeled samples for Delta-KNN (ablation) | Delta-KNN-RLog · ImageNet-1K, DFN CLIP | AURC (x1e-2): 3.98 -> 4.81 | No clear effect |
| use Delta-MDS-MSP instead of RLog combination on language task | Delta-MDS · Amazon Reviews in-distribution, DistilBERT | AURC (x1e-2): 12.4 -> 11.9 | Improved |
| use Energy score | Energy · ImageNet variants avg (1K), DFN CLIP | AURC (x1e-2): 11.5 -> 24.8 | Hurt |
Closest papers by meaning
- What Does It Take to Build a Performant Selective Classifier? (NeurIPS 2025 · similarity 0.89)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · similarity 0.86)
- SAND: One-Shot Feature Selection with Additive Noise Distortion (ICML 2025 · similarity 0.85)
- Overcoming Common Flaws in the Evaluation of Selective Classification Systems (NeurIPS 2024 · similarity 0.85)
- Rejection via Learning Density Ratios (NeurIPS 2024 · similarity 0.85)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · 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.
