Research map / Label-efficient and robust learning / OOD and anomaly detection
Optimal Parameter and Neuron Pruning for Out-of-Distribution Detection
Chao Chen, Zhihang Fu, Kai Liu, Ze Chen, Mingyuan Tao, Jieping Ye
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 |
|---|---|---|---|
| Prune parameters and neurons with unusually high or near-zero energy-score sensitivity | ResNet50 · ImageNet-1K, average | FPR95: 58.41 -> 25.93 | Improved |
| Add neuron pruning to parameter pruning (ablation) | ResNet50 · ImageNet-1K, average | FPR95: 34.00 -> 29.12 -> 25.93 | Improved |
| Add ReAct feature rectification to OPNP (ablation) | ResNet50 · ImageNet-1K, average | FPR95: 25.93 -> 23.12 | Improved |
| Use both parameter and neuron pruning rather than parameter pruning alone (ablation) | ViT-B/16 · ImageNet-1K, average | FPR95: 24.49 -> 25.19 | Mixed |
| Add OPNP to Energy | ResNet50 · CIFAR-10, average | FPR95: 9.14 -> 6.61 | Improved |
| Add OPNP to ReAct | ResNet50 · CIFAR-100, average | FPR95: 26.05 -> 23.56 | Mixed |
| Use sensitivity-guided parameter pruning rather than random or low-magnitude parameter pruning (ablation) | ResNet50 · ImageNet-1K / SUN | FPR95: 49.42, 43.36 -> 30.40 | Improved |
| Use sensitivity-guided neuron pruning rather than random or low-feature-norm neuron pruning (ablation) | ResNet50 · ImageNet-1K / SUN | FPR95: 52.44, 47.76 -> 26.67 | Improved |
| Apply OPNP to the FC layer rather than an intermediate convolutional layer (ablation) | ResNet50 · ImageNet-1K / SUN | FPR95: 54.98, 54.30, 46.51, 37.06 -> 18.50 | Improved |
| Use FC-only pruning rather than global threshold or layer-wise pruning (ablation) | ResNet50 · ImageNet-1K / SUN | FPR95: 39.74, 40.48 -> 18.50 | Improved |
| Estimate sensitivity from 1% rather than 100% of training samples (ablation) | OPP · ImageNet-1K / SUN | FPR95: 30.40 -> 32.57 | Hurt |
| Use variance rather than mean to measure neuron sensitivity (ablation) | OPNP · ImageNet-1K / SUN | FPR95: 26.7 -> 46.1 | Hurt |
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.
- ReAct — ReAct: Out-of-distribution Detection With Rectified Activations (NeurIPS 2021) · better on 4 of 4 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
- Enhancing the Power of OOD Detection via Sample-Aware Model Selection (CVPR 2024 · similarity 0.90)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.90)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.90)
- Double Descent Meets Out-of-Distribution Detection: Theoretical Insights and Empirical Analysis on the Role of Model Complexity (NeurIPS 2025 · similarity 0.90)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.90)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.89)
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.
