Research map / Label-efficient and robust learning / OOD and anomaly detection
Rethinking Reconstruction Autoencoder-Based Out-of-Distribution Detection
Yibo Zhou
Check 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 |
|---|---|---|---|
| Replace image reconstruction targets with classifier activation-vector features (ablation) | LSA OoD detector · CIFAR-100 vs. TinyImagenet-crop | FPR@95%TPR: 42.0 -> 99.5 | Hurt |
| Replace L2 distance with normalized L2 distance (ablation) | LSA with AV feature reconstruction · CIFAR-100 vs. TinyImagenet-crop | FPR@95%TPR: 99.5 -> 5.9 | Improved |
| Replace autoregressive latent-space regularization with cross-entropy regularization (ablation) | LSA with normalized L2 · CIFAR-100 vs. TinyImagenet-crop | FPR@95%TPR: 5.9 -> 3.0 | Improved |
| Use layerwise feature reconstruction instead of basic input reconstruction scoring (ablation) | LSA with cross-entropy regularizer · CIFAR-100 vs. TinyImagenet-crop | AUROC: 99.1 -> 91.3 | Hurt |
| Add epsilon terms to the layerwise certainty score (ablation) | Layerwise semantic reconstruction · CIFAR-100 vs. TinyImagenet-crop | AUROC: 91.3 -> 99.4 | Improved |
| Add epsilon terms to the layerwise certainty score (ablation) | Layerwise semantic reconstruction · CIFAR-100 vs. TinyImagenet-crop | FPR@95%TPR: 0.2 -> 1.5 | Hurt |
| Remove weight decay when training the classifier used for activation features (ablation) | Layerwise semantic reconstruction with Dense-BC · CIFAR-100 vs. TinyImagenet-crop | AUROC: 96.1 / 97.1 | Improved |
| Remove weight decay from Dense-BC classifier training (ablation) | Dense-BC classifier · CIFAR-100 classification | accuracy: 77% to 72% | Hurt |
| Remove weight decay from Wide-ResNet classifier training (ablation) | Wide-ResNet classifier · CIFAR-100 classification | accuracy: 78% to 77% | Hurt |
Closest papers by meaning
- Diffusion-based Layer-wise Semantic Reconstruction for Unsupervised Out-of-Distribution Detection (NeurIPS 2024 · similarity 0.87)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.84)
- Distributional Autoencoders Know the Score (NeurIPS 2025 · similarity 0.84)
- Outlier-aware Slicing for Post-Training Quantization in Vision Transformer (ICML 2024 · similarity 0.83)
- Out-of-Distribution Detection with Prototypical Outlier Proxy (AAAI 2025 · similarity 0.83)
- Image-to-video Adaptation with Outlier Modeling and Robust Self-learning (AAAI 2025 · 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.
