Research map / Label-efficient and robust learning / OOD and anomaly detection
CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances
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 |
|---|---|---|---|
| Train with CSI rather than the strongest listed baseline | One-class detection · CIFAR-10 | AUROC: 89.8 -> 94.3 | Helped |
| Train with CSI rather than the strongest listed baseline | One-class detection · CIFAR-100 (super-class) | AUROC: 78.7 -> 89.6 | Helped |
| Train with CSI rather than the strongest listed baseline | One-class detection · ImageNet-30 | AUROC: 85.7 -> 91.6 | Helped |
| Use CSI instead of Rot+Trans | CIFAR-10 OOD detection · CIFAR-10 → CIFAR-100 | AUROC: 82.3\stdv{0.2} -> 89.2\stdv{0.1} | Helped |
| Use CSI instead of Rot+Trans | ImageNet-30 OOD detection · ImageNet-30 → Places-365 | AUROC: 53.1\stdv{1.7} -> 78.3\stdv{0.3} | Helped |
| Add contrastive shifted-instance loss to SimCLR training (ablation) | CSI representation · One-class CIFAR-10 | AUROC: 87.9 -> 91.6 | Helped |
| Add shifted-instance classification to the con-SI objective (ablation) | CSI representation · One-class CIFAR-10 | AUROC: 91.6 -> 94.3 | Helped |
| Use shifted-instance classification alone instead of the full CSI objective (ablation) | CSI representation · One-class CIFAR-10 | AUROC: 94.3 -> 88.6 | Hurt |
| Add shifted-instance contrast score and ensembling to the contrastive score (ablation) | CSI detection score · One-class CIFAR-10 | AUROC: 91.3 -> 93.3 | Helped |
| Use the combined contrastive and shifted-instance classification scores (ablation) | CSI detection score · One-class CIFAR-10 | AUROC: 93.8 -> 94.3 | Helped |
| Use Gaussian noise instead of rotation as the shifting transformation (ablation) | CSI · DTD → Textile | AUROC: 65.9 -> 80.1 | Helped |
| Ensemble predictions over shifting transformations (ablation) | CSI classifier · Labeled CIFAR-10 | test accuracy: 94.8\stdv{0.1} -> 96.1\stdv{0.1} | Helped |
Closest papers by meaning
- Universal Novelty Detection Through Adaptive Contrastive Learning (CVPR 2024 · similarity 0.90)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.84)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.84)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.83)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.83)
- Contrastive Mean-Shift Learning for Generalized Category Discovery (CVPR 2024 · similarity 0.83)
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.