Research map / Label-efficient and robust learning / OOD and anomaly detection
DICE: Leveraging Sparsification for Out-of-Distribution Detection
Yiyou Sun, Yixuan Li
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 DICE weight sparsification to the energy score | Energy score · ImageNet-1k ID, avg of 4 OOD sets | FPR95: 58.41 -> 34.75 | Improved |
| add DICE weight sparsification to the energy score | Energy score · CIFAR-100 ID, avg of 6 OOD sets | FPR95: 68.45 -> 49.72 | Improved |
| add DICE weight sparsification to the energy score | Energy score · CIFAR-10 ID, avg of 6 OOD sets | FPR95: 26.55 -> 20.83 | Improved |
| add DICE weight sparsification on top of ReAct | ReAct · ImageNet-1k ID, avg of 4 OOD sets | FPR95: 31.43 -> 27.25 | Improved |
| use directed (contribution-based) masking instead of random weight dropout (ablation) | DICE · ImageNet-1k ID, avg of 4 OOD sets | FPR95: 76.28 -> 34.75 | Improved |
| use contribution-directed masking instead of L1 weight pruning (ablation) | DICE · ImageNet-1k ID, avg of 4 OOD sets | FPR95: 52.81 -> 34.75 | Improved |
| use contribution-directed masking instead of random unit dropout (ablation) | DICE · ImageNet-1k ID, avg of 4 OOD sets | FPR95: 83.91 -> 34.75 | Improved |
| use contribution-directed masking instead of L2 unit pruning (ablation) | DICE · ImageNet-1k ID, avg of 4 OOD sets | FPR95: 90.80 -> 34.75 | Improved |
| keep top-k contributions instead of random-k (ablation) | DICE · CIFAR-100 ID, avg of 6 OOD sets | FPR95: 77.48 -> 49.72 | Improved |
| keep top-k contributions instead of bottom-k (ablation) | DICE · CIFAR-100 ID, avg of 6 OOD sets | FPR95: 99.70 -> 49.72 | Improved |
| keep top-k contributions instead of top+bottom-k (ablation) | DICE · CIFAR-10 ID, avg of 6 OOD sets | FPR95: 24.25 -> 20.83 | Improved |
| add mild sparsity p=0.1 versus none (p=0) (ablation) | DICE · ImageNet-1k ID, avg of 4 OOD sets, sparsity sweep | FPR95: 58.41 -> 40.25 | Improved |
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
- Fast Decision Boundary based Out-of-Distribution Detector (ICML 2024 · similarity 0.90)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.89)
- Double Descent Meets Out-of-Distribution Detection: Theoretical Insights and Empirical Analysis on the Role of Model Complexity (NeurIPS 2025 · similarity 0.89)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.89)
- What If the Input is Expanded in OOD Detection? (NeurIPS 2024 · similarity 0.89)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · 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.
