Research map / Learning theory and algorithms / Matrix and tensor methods
Robustness Can Be Cheap: A Highly Efficient Approach to Discover Outliers under High Outlier Ratios
Siqi Wang, En Zhu, Xiping Hu, Xinwang Liu, Qiang Liu, Jianping Yin, Fei Wang
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 LEOD-basic/fast low-rank framework instead of other outlier detectors (best other: RKDE) | LEOD / UOCL · SVHN under high outlier ratio | f1-score: at most 29.5% f1-score gain | Improved |
| Use low-rank similarity (LEOD) instead of full similarity matrix (UOCL) | LEOD-basic · Caltech101, outlier ratio 0.6 | f1-score: 82.6 -> 85.3 | Improved |
| Use LEOD-fast instead of UOCL | LEOD-fast · SVHN, outlier ratio 0.6 | time (sec.): 3439.0 -> 3.4 | Improved |
| Use LEOD-basic instead of UOCL | LEOD-basic · SVHN, outlier ratio 0.6 | peak memory (GiB): 39.06 -> 0.36 | Improved |
| Use LEOD-fast instead of LEOD-basic (closed-form alpha update) | LEOD · SVHN, outlier ratio 0.6 | time (sec.): 247.2 -> 3.4 | Improved |
| Use LEOD-basic instead of LEOD-fast | LEOD · SVHN, outlier ratio 0.6 | peak memory (GiB): 0.40 -> 0.36 | Improved |
| Naively low-rank approximate (I + gamma1 L) instead of sparsifying L (ablation) | LEOD · Caltech101 / Cifar-10 | f1-score: at most 6% performance loss | Hurt |
| Vary lambda from 0.01 to 10.0 (ablation) | LEOD-fast · Caltech101 | f1-score | No clear effect |
| Large lambda = 100 at outlier ratio 60% (ablation) | LEOD-fast · Caltech101, outlier ratio 0.6 | f1-score: at most 4% f1 gain | Improved |
| Large lambda = 100 at outlier ratio <= 40% (ablation) | LEOD-fast · Caltech101 | f1-score | Hurt |
| Very small lambda = 0.01 at outlier ratio 60% (ablation) | LEOD-fast · Caltech101, outlier ratio 0.6 | f1-score | Hurt |
| Vary m from n/5 to n/20 and l from n/10 to n/30 (ablation) | LEOD · Caltech101 | f1-score: variance within 3% | No clear effect |
Closest papers by meaning
- ALTBI: Constructing Improved Outlier Detection Models via Optimization of Inlier-Memorization Effect (AAAI 2025 · similarity 0.86)
- Density Ratio Estimation with Doubly Strong Robustness (ICML 2024 · similarity 0.85)
- Robust Conformal Outlier Detection under Contaminated Reference Data (ICML 2025 · similarity 0.85)
- Tight Rates in Supervised Outlier Transfer Learning (ICLR 2024 · similarity 0.84)
- Stray Intrusive Outliers-Based Feature Selection on Intra-Class Asymmetric Instance Distribution or Multiple High-Density Clusters (ICML 2025 · similarity 0.84)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.84)
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.
