Research map / Label-efficient and robust learning / OOD and anomaly detection
Detecting Out-of-Distribution Examples with Gram Matrices
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 |
|---|---|---|---|
| Gram matrix anomaly detection (all orders, all layers, min/max deviation, normalized) vs. baselines | far-OOD detection with pretrained ResNet/DenseNet · CIFAR-10/CIFAR-100/SVHN in-distribution, far-OOD sets | TNR at TPR 95% | Helped |
| Gram matrix detector applied to near-OOD (CIFAR-10 vs CIFAR-100) | ResNet34 pretrained without OE · CIFAR-10 vs CIFAR-100 | TNR at TPR 95%: 19.1 vs 20.2 for Mahalanobis | Hurt |
| Adding Gram matrix deviation on top of Outlier Exposure training (Ours + MSP) | WRN 40-2 trained with OE · CIFAR-10 vs iSUN and other OOD sets | TNR at TPR95: 99.8 vs 98.3 for MSP alone on iSUN | Helped |
| Using only lower-level blocks vs. deeper blocks for Gram features (ablation) | Gram matrix detector · ResNet/DenseNet, CIFAR vs Tiny ImageNet | TNR at TPR95 | Mixed |
| Using higher-order Gram matrices (up to p=10) instead of only low orders (ablation) | Gram matrix detector · ResNet/DenseNet, CIFAR vs Tiny ImageNet | TNR at TPR95 | Helped |
| Normalizing layerwise deviations by expected validation deviation vs. no normalization (ablation) | Gram matrix detector · ablation over Q1-Q3 | TNR at TPR95 | Helped |
| Deviation from class-conditional mean instead of extrema (ablation) | Gram matrix detector · SVHN vs CIFAR-10 | TNR at TPR95: improves SVHN vs CIFAR-10 to 94% but overall extrema is strongest | Mixed |
Closest papers by meaning
- Out-of-Distribution Detection based on In-Distribution Data Patterns Memorization with Modern Hopfield Energy (ICLR 2023 · similarity 0.89)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.88)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.88)
- Exploiting Discrepancy in Feature Statistic for Out-of-Distribution Detection (AAAI 2024 · similarity 0.87)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.87)
- ImageNet-OOD: Deciphering Modern Out-of-Distribution Detection Algorithms (ICLR 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.