Research map / Label-efficient and robust learning / OOD and anomaly detection
Backpropagated Gradient Representations for Anomaly Detection
Gukyeong Kwon, Mohit Prabhushankar, Dogancan Temel, Ghassan AlRegib
Read the paperCheck 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 |
|---|---|---|---|
| add gradient constraint and score with gradient loss (ablation) | CAE · CIFAR-10 (one-class, avg AUROC) | AUROC: 0.564 -> 0.661 | Improved |
| add gradient constraint (reconstruction score) (ablation) | CAE · CIFAR-10 (one-class, avg AUROC) | AUROC: 0.564 -> 0.554 | No clear effect |
| add gradient constraint and score with gradient loss (ablation) | VAE · CIFAR-10 (one-class, avg AUROC) | AUROC: 0.583 -> 0.647 | Improved |
| add gradient constraint (latent loss score) (ablation) | VAE · CIFAR-10 (one-class, avg AUROC) | AUROC: 0.583 -> 0.550 | Hurt |
| add latent (KL) constraint to CAE, reconstruction score (ablation) | CAE · CIFAR-10 (one-class, avg AUROC) | AUROC: 0.564 -> 0.526 | Hurt |
| use all decoder layers' gradients instead of first layer (ablation) | GradCon · CURE-TSR Exposure | AUROC: 0.859 -> 0.891 | Improved |
| use all decoder layers' gradients instead of 4th layer only (ablation) | GradCon · CIFAR-10 (gradient loss per decoder layer) | AUROC: 0.605 -> 0.661 | Improved |
| combine reconstruction error with gradient loss as score (GradCon vs Grad only) (ablation) | GradCon · fMNIST (10% outliers) | AUC: 0.931 -> 0.938 | Improved |
| combine reconstruction error with gradient loss as score (GradCon vs Grad only) (ablation) | GradCon · fMNIST (50% outliers) | F1: 0.870 -> 0.871 | No clear effect |
| increase gradient-loss weight beta up to 2 alpha (ablation) | GradCon · CIFAR-10 (one-class, avg AUROC) | AUROC | Improved |
| replace adversarial/autoregressive models with CAE plus gradient loss | GradCon · CIFAR-10 (one-class, avg AUROC) | AUROC: 0.657 -> 0.664 | Improved |
| replace adversarial/autoregressive models with CAE plus gradient loss | GradCon · MNIST (one-class, avg AUROC) | AUROC: 0.975 -> 0.973 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OCGAN — OCGAN: One-Class Novelty Detection Using GANs With Constrained Latent Representations (CVPR 2019) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.88)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.87)
- Old Is Gold: Redefining the Adversarially Learned One-Class Classifier Training Paradigm (CVPR 2020 · similarity 0.87)
- ResAD++: Towards Class Agnostic Anomaly Detection via Residual Feature Learning (IJCV 2026 · similarity 0.87)
- Scalable, Explainable and Provably Robust Anomaly Detection with One-Step Flow Matching (NeurIPS 2025 · similarity 0.85)
- Contamination-Resilient Anomaly Detection via Adversarial Learning on Partially-Observed Normal and Anomalous Data (ICML 2024 · 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.
