Research map / Label-efficient and robust learning / OOD and anomaly detection
Kernel PCA for Out-of-Distribution Detection: Non-Linear Kernel Selection and Approximation
Kun Fang, Qinghua Tao, Mingzhen He, Kexin Lv, Runze Yang, Haibo Hu, Xiaolin Huang, Jie Yang, Longbing Cao
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 the Cosine kernel instead of linear PCA (ablation) | KPCA (random-feature/Nystrom framework) · ImageNet-1K ResNet50 (avg) | FPR95: 84.41 -> 58.40 | Improved |
| use the Cosine kernel instead of linear PCA (ablation) | KPCA · ImageNet-1K ResNet50 (avg) | AUROC: 57.46 -> 83.86 | Improved |
| add the Cosine kernel in front of the Gaussian kernel (random features) (ablation) | KPCA_rff · ImageNet-1K ResNet50 (avg) | FPR95: 94.88 -> 48.60 | Improved |
| add the Cosine kernel in front of the Gaussian kernel (Nystrom) (ablation) | KPCA_nys · ImageNet-1K ResNet50 (avg) | AUROC: 35.44 -> 91.02 | Improved |
| add a Gaussian kernel on top of the Cosine kernel (Nystrom) (ablation) | KPCA_nys · ImageNet-1K ResNet50 (avg) | FPR95: 58.40 -> 38.15 | Improved |
| add a Laplacian kernel on top of the Cosine kernel (random features) (ablation) | KPCA_rff · ImageNet-1K ResNet50 (avg) | FPR95: 58.40 -> 62.54 | Hurt |
| add a Polynomial kernel on top of the Cosine kernel (random features) (ablation) | KPCA_rff · ImageNet-1K ResNet50 (avg) | FPR95: 58.40 -> 68.01 | Hurt |
| replace random Fourier features with the data-dependent Nystrom approximation (ablation) | Cosine-Gaussian KPCA · ImageNet-1K ResNet50 (avg) | FPR95: 48.60 -> 38.15 | Improved |
| sample low-energy InD landmarks instead of uniform for Nystrom (ablation) | KPCA_nys · ImageNet-1K ResNet50 (avg) | FPR95: 49.94 -> 38.15 | Improved |
| sample low-energy landmarks instead of high-energy ones (ablation) | KPCA_nys · ImageNet-1K ResNet50 (avg) | FPR95: 58.23 -> 38.15 | Improved |
| apply KPCA on layer4 features instead of layer3 (ablation) | KPCA_nys · ImageNet-1K ResNet50 (avg) | FPR95: 61.59 -> 38.15 | Improved |
| add ReAct feature rectification | KPCA_nys · ImageNet-1K ResNet50 (avg) | FPR95: 38.15 -> 15.26 | Improved |
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.
- GradOrth — GradOrth: A Simple yet Efficient Out-of-Distribution Detection with Orthogonal Projection of Gradients (NeurIPS 2023) · 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
- Kernel PCA for Out-of-Distribution Detection (NeurIPS 2024 · similarity 0.96)
- GradPCA: Leveraging NTK Alignment for Reliable Out-of-Distribution Detection (ICLR 2026 · similarity 0.90)
- How to Overcome Curse-of-Dimensionality for Out-of-Distribution Detection? (AAAI 2024 · similarity 0.89)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.88)
- Exploiting Discrepancy in Feature Statistic for Out-of-Distribution Detection (AAAI 2024 · similarity 0.87)
- Test-Time Linear Out-of-Distribution Detection (CVPR 2024 · similarity 0.87)
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.
