Research map / Label-efficient and robust learning / OOD and anomaly detection
A Unified Interpretation of Training-Time Out-of-Distribution Detection
Xu Cheng, Xin Jiang, Zechao Li
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 |
|---|---|---|---|
| penalize high-order interactions so model encodes mainly low-order ones (vs cross-entropy baseline) (ablation) | ResNet-18 · CIFAR-10 ID, OOD average | FPR95: 62.03 -> 91.45 | Hurt |
| penalize high-order interactions so model encodes mainly low-order ones (ablation) | ResNet-18 · CIFAR-10 ID, OOD average | AUROC: 88.48 -> 73.07 | Hurt |
| penalize high-order interactions so model encodes mainly low-order ones (ablation) | ResNet-34 · CIFAR-100 ID, OOD average | AUROC: 78.30 -> 58.77 | Hurt |
| penalize high-order interactions so model encodes mainly low-order ones (ablation) | WideResNet-40-2 · CIFAR-100 ID, OOD average | FPR95: 78.01 -> 90.33 | Hurt |
| penalize low-order interactions so model encodes mainly high-order ones (ablation) | ResNet-18 · CIFAR-10 ID, OOD average | FPR95: 62.03 -> 53.05 | Improved |
| penalize low-order interactions so model encodes mainly high-order ones (ablation) | ResNet-18 · CIFAR-100 ID, OOD average | AUROC: 78.15 -> 79.45 | Improved |
| penalize low-order interactions so model encodes mainly high-order ones (ablation) | WideResNet-40-2 · CIFAR-10 ID, OOD average | FPR95: 56.99 -> 61.64 | Hurt |
| penalize low-order interactions so model encodes mainly high-order ones (ablation) | ResNet-34 · CIFAR-100 ID, OOD average | FPR95: 78.95 -> 81.51 | Hurt |
| train with training-time OOD methods (CSI, LogitNorm, T2FNorm, DAL) instead of cross-entropy only | ResNet-18/34, WideResNet-40-2 · CIFAR-10 and CIFAR-100 ID | relative interaction strength: Delta R(m) > 0 for m > 0.75n, < 0 for m < 0.25n | Improved |
| train with training-time OOD methods instead of cross-entropy only | ResNet-based models · near/far-OOD interaction similarity | similarity of interaction distribution to ID | Improved |
Closest papers by meaning
- A Statistical Framework for Efficient Out of Distribution Detection in Deep Neural Networks (ICLR 2022 · similarity 0.91)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.90)
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.89)
- Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection (NeurIPS 2025 · similarity 0.89)
- Fast Decision Boundary based Out-of-Distribution Detector (ICML 2024 · similarity 0.89)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.89)
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.
