Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning and Evaluating Representations for Deep One-Class Classification
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 |
|---|---|---|---|
| Replace the rotation-prediction classifier with KDE on the learned representation (ablation) | Rotation-prediction one-class classifier · Five-dataset mean | AUC: 83.1 -> 86.6 | Helped |
| Add distribution augmentation to contrastive representation learning (ablation) | Contrastive learning · Five-dataset mean | AUC: 86.9 -> 89.9 | Helped |
| Use a moderate contrastive batch size of 32 rather than large batches (ablation) | Contrastive learning · CIFAR-10 validation | AUC: null | Helped |
| Use the encoder representation f rather than the projection-head output g∘f for evaluation (ablation) | Contrastive representations · CIFAR-10 | AUC: null | Helped |
| Add an MLP projection head during rotation-prediction training (ablation) | Rotation-prediction representation · Five-dataset mean | AUC: 86.6 -> 88.2 | Helped |
| Use a rotation classifier instead of KDE on the same RotNet representation (ablation) | RotNet representation · MVTec AD | Detection AUC: 83.5{ $\pm3.0$} -> 71.0{ $\pm3.5$} | Hurt |
| Add distribution augmentation to vanilla contrastive learning (ablation) | Contrastive representations · MVTec AD | Detection AUC: 80.2{ $\pm1.8$} -> 86.5{ $\pm1.6$} | Helped |
| Use larger contrastive batch sizes that make representations more uniform (ablation) | Contrastive learning · CIFAR-10 | AUC: null | Hurt |
| Train rotation prediction on randomly rotated CIFAR-10 images (ablation) | Rotation-prediction representation · CIFAR-10 | AUC: 60.7 | Hurt |
| Use rotation-augmented contrastive learning rather than rotation prediction under random-rotation training data (ablation) | Contrastive representation · CIFAR-10 | AUC: 92.4 | Helped |
| Ensemble five models trained with different distribution augmentations (ablation) | Contrastive (DA) · CIFAR-10 | AUC: 93.7{ $\pm0.6$} -> 94.6 | Helped |
Closest papers by meaning
- Rotation Has Two Sides: Evaluating Data Augmentation for Deep One-class Classification (ICLR 2024 · similarity 0.87)
- CLEAR: Cumulative LEARning for One-Shot One-Class Image Recognition (CVPR 2018 · similarity 0.86)
- One-Shot Exemplars for Class Grounding in Self-Supervised Learning (ICLR 2026 · similarity 0.86)
- Deep One-Class Classification via Interpolated Gaussian Descriptor (AAAI 2022 · similarity 0.86)
- I-Con: A Unifying Framework for Representation Learning (ICLR 2025 · similarity 0.85)
- Measuring Self-Supervised Representation Quality for Downstream Classification Using Discriminative Features (AAAI 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.