Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Uncertainty Attribution-Guided Fine-Tuning for Open-Set Action Recognition
Shehan Senavirathna, Hongji Guo, Qiang Ji
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 |
|---|---|---|---|
| apply Stage 1-3 refinement to deep ensemble | Vanilla DE · UCF101 vs HMDB51 (all four backbones) | AUROC: TPN 84.39 -> 88.62; TSM 81.83 -> 85.51; I3D 85.10 -> 87.95 | Improved |
| add Stage 3 UAM-IA to Stages 1+2 (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 86.21 -> 88.62 | Improved |
| add Stage 1 motion guidance on top of Stages 2+3 (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 87.76 -> 88.62 | Improved |
| Stage 1 alone vs vanilla ensemble (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 84.39 -> 84.78 | Improved |
| add Stage 2 epistemic regularization (Stage 1 + 3 present) (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 88.03 -> 88.62 | Improved |
| regularize epistemic uncertainty instead of aleatoric (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 85.38 -> 86.21 | Improved |
| regularize total uncertainty instead of CE only (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 85.43 -> 85.94 | Improved |
| regularize aleatoric uncertainty instead of CE only (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 85.43 -> 85.38 | No clear effect |
| UAM-IA instead of frame-difference mask (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 87.18 -> 88.62 | Improved |
| amplify low-attribution regions (UAM-IA) vs suppress high-attribution (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 87.42 -> 88.62 | Improved |
| increase UAM-IA alpha from 1.00 to 2.00 (ablation) | ensemble · UCF101 vs HMDB51 (TPN) | AUROC: 88.62 -> 87.26 | Hurt |
| distilled student vs SOAR (single model) | student · UCF101 vs MiT-v2 / HMDB51 | AUROC: TPN HMDB 86.67 -> 88.50; TSM HMDB 85.96 -> 84.24; I3D MiT 94.60 -> 93.98 | Mixed |
Closest papers by meaning
- Uncertainty-Aware Dual-Evidential Learning for Weakly-Supervised Temporal Action Localization (TPAMI 2023 · similarity 0.88)
- Evidential Deep Learning for Open Set Action Recognition (ICCV 2021 · similarity 0.87)
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · similarity 0.85)
- ExACT: Language-guided Conceptual Reasoning and Uncertainty Estimation for Event-based Action Recognition and More (CVPR 2024 · similarity 0.85)
- VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention (AAAI 2025 · similarity 0.85)
- Storyboard-guided Alignment for Fine-grained Video Action Recognition (NeurIPS 2025 · similarity 0.84)
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.
