Research map / Label-efficient and robust learning / OOD and anomaly detection
POEM: Out-of-Distribution Detection with Posterior Sampling
Yifei Ming, Ying Fan, 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 |
|---|---|---|---|
| replace greedy outlier mining with posterior sampling | OOD detection with auxiliary outlier data · CIFAR-10; six OOD test datasets | FPR95: 4.38 -> 2.54 | Improved |
| replace greedy outlier mining with posterior sampling | OOD detection with auxiliary outlier data · CIFAR-100; six OOD test datasets | FPR95: 19.96 -> 15.14 | Improved |
| replace random outlier sampling with posterior-sampled outlier mining | energy-regularized OOD detection · CIFAR-100; six OOD test datasets | FPR95: 19.25 -> 15.14 | Improved |
| add energy-regularized training instead of the K+1-class training scheme (ablation) | POEM · CIFAR-10; six natural OOD test datasets | FPR95: 2.81 -> 2.54 | Improved |
| add energy-regularized training instead of the K+1-class training scheme (ablation) | POEM · CIFAR-100; six natural OOD test datasets | FPR95: 17.11 -> 15.14 | Improved |
| decrease the auxiliary pool size by half (ablation) | POEM · CIFAR-10; six OOD test datasets | FPR95: 2.54 -> 4.23 | Hurt |
| decrease the auxiliary pool size by half (ablation) | POEM · CIFAR-10; six OOD test datasets | AUROC: 99.40 -> 99.29 | Hurt |
| stop outlier mining and posterior updates early while continuing regular training (ablation) | POEM · CIFAR-100; six OOD test datasets | AUROC: 97.79 -> 97.33\% | Hurt |
| use three times more randomly sampled outliers (ablation) | Energy with auxiliary outlier data · CIFAR-100; six OOD test datasets | FPR95: 19.25 -> 19.19 | No clear effect |
| use TinyImages as the auxiliary outlier dataset (ablation) | POEM · CIFAR-10; six OOD test datasets | FPR95: 3.27 -> 1.56 | Improved |
Later papers that compared with this method
6 later accepted papers report POEM as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · better on 1 of 3 shared results)
- Out-Of-Distribution Detection with Diversification (Provably) (NeurIPS 2024 · better on 2 of 2 shared results)
- Energy-based Hopfield Boosting for Out-of-Distribution Detection (NeurIPS 2024 · better on 2 of 2 shared results)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · better on 1 of 1 shared result)
- Learning to Augment Distributions for Out-of-distribution Detection (NeurIPS 2023 · better on 6 of 7 shared results)
- Diversified Outlier Exposure for Out-of-Distribution Detection via Informative Extrapolation (NeurIPS 2023 · better on 1 of 1 shared result)
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
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.90)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.89)
- Out-Of-Distribution Detection with Diversification (Provably) (NeurIPS 2024 · similarity 0.88)
- ODIM: Outlier Detection via Likelihood of Under-Fitted Generative Models (ICML 2024 · similarity 0.88)
- Out-of-Distribution Detection with Prototypical Outlier Proxy (AAAI 2025 · similarity 0.88)
- Energy-based Hopfield Boosting for Out-of-Distribution Detection (NeurIPS 2024 · similarity 0.88)
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.
