Research map / Label-efficient and robust learning / OOD and anomaly detection
Task-Agnostic Undesirable Feature Deactivation Using Out-of-Distribution Data
Dongmin Park, Hwanjun Song, Minseok Kim, Jae-Gil Lee
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 |
|---|---|---|---|
| replace softmax-level calibration (OAT) with feature-level calibration | OAT · CIFAR-10 N=500 (LSUN) | accuracy: 40.73 -> 42.51 | Improved |
| add LSUN OOD data instead of SVHN | TAUFE · CIFAR-10 N=500 | accuracy: 41.58 -> 42.51 | Improved |
| add Places365 instead of ImageNet-990 as OOD (small data) | TAUFE · ImageNet-10 64x64 N=500 | accuracy: 42.80 -> 43.25 | Mixed |
| add TAUFE regularization to MixMatch | MixMatch · CIFAR-10 semi-supervised | accuracy: 88.32 -> 90.10 | Improved |
| add TAUFE regularization to MixMatch | MixMatch · CIFAR-100 semi-supervised | accuracy: 51.38 -> 52.58 | Improved |
| add TAUFE with LSUN OOD to PGD adversarial training | PGD · CIFAR-10 | adversarial accuracy: 44.26 -> 45.48 | Improved |
| add TAUFE with LSUN OOD to PGD adversarial training | PGD · CIFAR-10 | clean accuracy: 71.22 -> 72.31 | Improved |
| add TAUFE feature-level calibration for box regression | Standard ResNet-50 · CAR N=2,000 L1-IoU | IoU: 83.52 -> 85.82 | Improved |
| use ImageNet instead of Places365 as OOD | TAUFE · CUB200 box regression | IoU: 77.24 -> 77.01 | Improved |
| replace softmax-level with feature-level calibration for WSOL | OAT · CUB200 N=2,000 (ImageNet OOD) | GT-known Loc: 55.24 -> 59.68 | Improved |
| increase image resolution of ImageNet-10 from 64 to 224 | TAUFE · ImageNet-10 | accuracy | No clear effect |
| add TAUFE to OOD detection | TAUFE · OOD detection | detection performance | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MixMatch — MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Understanding the Feature Norm for Out-of-Distribution Detection (ICCV 2023 · similarity 0.84)
- Removing Undesirable Feature Contributions Using Out-of-Distribution Data (ICLR 2021 · similarity 0.84)
- Understanding and Improving Feature Learning for Out-of-Distribution Generalization (NeurIPS 2023 · similarity 0.84)
- CADRef: Robust Out-of-Distribution Detection via Class-Aware Decoupled Relative Feature Leveraging (CVPR 2025 · similarity 0.84)
- Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimation (ICML 2024 · similarity 0.83)
- Feature Contamination: Neural Networks Learn Uncorrelated Features and Fail to Generalize (ICML 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.
