Research map / Label-efficient and robust learning / OOD and anomaly detection
On the Value of Out-of-Distribution Testing: An Example of Goodhart's Law
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| invert training answer distribution for sampling (exploit OOD split construction) | random predictions · VQA-CP test yes/no | accuracy (%): 25.87 -> 83.25 | Helped |
| invert training answer distribution for sampling | random predictions · VQA-CP val yes/no (in-domain) | accuracy (%): 70.10 -> 55.36 | Hurt |
| mask top predicted answer at test time | BUTD · VQA-CP test yes/no | accuracy (%): 42.98 -> 82.44 | Helped |
| mask top predicted answer at test time | BUTD · VQA-CP val overall (in-domain) | accuracy (%): 64.73 -> 30.90 | Hurt |
| add random-image regularizer lambda=12 | BUTD · VQA-CP test overall | accuracy (%): 38.82 -> 55.37 | Helped |
| add random-image regularizer lambda=12 | BUTD · VQA-CP val overall (in-domain) | accuracy (%): 64.73 -> 54.24 | Hurt |
| add random-image regularizer lambda=5 | BUTD · VQA-CP test other | accuracy (%): 43.95 -> 45.99 | Helped |
| add random-image regularizer lambda=12 then evaluate after retraining on VQA v2 | BUTD · VQA v2 val overall | accuracy (%): 59.93 -> 57.24 | Hurt |
| choose training-set label balance (VQA-CP train vs VQA v2 retrain) for in-domain evaluation | random predictions, inverted · in-domain yes/no | accuracy (%): 55.36 (VQA-CP val) vs 64.11 (VQA v2 val) | Mixed |
Closest papers by meaning
- OODD: Test-time Out-of-Distribution Detection with Dynamic Dictionary (CVPR 2025 · similarity 0.84)
- Test-Time Linear Out-of-Distribution Detection (CVPR 2024 · similarity 0.83)
- What If the Input is Expanded in OOD Detection? (NeurIPS 2024 · similarity 0.83)
- A Noisy Elephant in the Room: Is Your Out-of-Distribution Detector Robust to Label Noise? (CVPR 2024 · similarity 0.83)
- Out-of-Distribution Detection with Prototypical Outlier Proxy (AAAI 2025 · similarity 0.82)
- Rethinking Out-of-Distribution Detection on Imbalanced Data Distribution (NeurIPS 2024 · similarity 0.82)
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.