Research map / Probabilistic and scientific modeling / Variational and generative sampling
Generative Probabilistic Novelty Detection with Adversarial Autoencoders
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 |
|---|---|---|---|
| Use GPND probabilistic scoring instead of the strongest listed non-GPND baseline | MNIST novelty detection · MNIST, 10%–50% outliers | F1: 0.97 -> 0.983; 0.92 -> 0.971; 0.92 -> 0.961; 0.91 -> 0.950; 0.88 -> 0.939 | Helped |
| Use GPND instead of R-graph | Coil-100 novelty detection · Coil-100, one, four, or seven inlier categories | F1: 0.990 -> 0.979; 0.970 -> 0.960; 0.955 -> 0.941 | Hurt |
| Use GPND instead of ODIN-Dense-BC | CIFAR-10 novelty detection · CIFAR-10, TinyImageNet (crop) | AUROC: 99.1 -> 90.1 | Hurt |
| Use GPND instead of ODIN-Dense-BC | CIFAR-100 novelty detection · CIFAR-100, TinyImageNet (resize) and LSUN (resize) | AUROC: 90.7 -> 95.9; 91.5 -> 98.3 | Helped |
| Replace the AAE with a VAE (ablation) | GPND · MNIST, 10%–50% outliers | F1: 98.2 -> 97.6; 97.1 -> 95.8; 96.1 -> 94.2; 95.0 -> 92.4; 93.9 -> 90.5 | Hurt |
| Remove the adversarial discriminator applied to decoded images (ablation) | GPND · MNIST, 10%–50% outliers | F1: 98.2 -> 97.3; 97.1 -> 95.5; 96.1 -> 94.0; 95.0 -> 92.0; 93.9 -> 90.2 | Hurt |
| Use reconstruction-error thresholding instead of GPND scoring (ablation) | GPND · MNIST, 10%–50% outliers | F1: 98.2 -> 84.8; 97.1 -> 79.6; 96.1 -> 79.5; 95.0 -> 77.6; 93.9 -> 75.6 | Hurt |
| Drop the perpendicular probability component (ablation) | GPND full probability score · MNIST | F1 | Hurt |
| Drop the parallel probability component (ablation) | GPND full probability score · MNIST | F1 | Hurt |
| Drop the Jacobian scaling factor from the parallel component (ablation) | GPND parallel-only score · MNIST | F1 | Hurt |
| Vary latent-space size from 16 to 64 (ablation) | GPND · MNIST and Coil-100 | F1 | No clear effect |
Closest papers by meaning
- Adversarially Learned One-Class Classifier for Novelty Detection (CVPR 2018 · similarity 0.90)
- Fake It Until You Make It : Towards Accurate Near-Distribution Novelty Detection (ICLR 2023 · similarity 0.89)
- Universal Novelty Detection Through Adaptive Contrastive Learning (CVPR 2024 · similarity 0.87)
- ODIM: Outlier Detection via Likelihood of Under-Fitted Generative Models (ICML 2024 · similarity 0.85)
- Contamination-Resilient Anomaly Detection via Adversarial Learning on Partially-Observed Normal and Anomalous Data (ICML 2024 · similarity 0.85)
- ALTBI: Constructing Improved Outlier Detection Models via Optimization of Inlier-Memorization Effect (AAAI 2025 · similarity 0.85)
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.