Research map / Reinforcement learning / Imitation learning
ERA: Expert Retrieval and Assembly for Early Action Prediction
Lin Geng Foo, Tianjiao Li, Hossein Rahmani, Qiuhong Ke, Jun Liu
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 ERA modules (expert retrieval and assembly) to the backbone (ablation) | 2s-AGCN baseline · NTU60 | AUC: 66.43 -> 71.62 | Improved |
| add Expert Learning Rate Optimization (ELRO) (ablation) | ERA-Net w/o ELRO · NTU60 | AUC: 71.62 -> 73.87 | Improved |
| add ELRO (ablation) | ERA-Net w/o ELRO · NTU60 (20% observed) | accuracy: 43.94 -> 53.98 | Improved |
| add ELRO (ablation) | ERA-Net w/o ELRO · UCF101 | AUC: 92.51 -> 93.64 | Improved |
| add ERA modules | 2s-AGCN baseline · NTU120 | AUC: 50.03 -> 57.02 | Improved |
| add ERA modules | 3D ResNeXt-101 baseline · UCF101 | AUC: 89.30 -> 93.64 | Improved |
| use 20:80 expert to non-expert split instead of all non-experts (0:100) (ablation) | ERA module · NTU60 | AUC: 66.43 -> 73.87 | Improved |
| use 20:80 split instead of 100:0 experts only (ablation) | ERA module · NTU60 | AUC: 70.12 -> 73.87 | Improved |
| increase Expert Bank size M from 2 to 5 (ablation) | ERA module · NTU60 | AUC: 71.55 -> 73.87 | Improved |
| increase Expert Bank size M from 5 to 10 (ablation) | ERA module · NTU60 | AUC: 73.87 -> 73.86 | No clear effect |
| replace more convolutional layers with ERA (25% to 100%) (ablation) | ERA-Net · NTU60 | AUC: 73.87 -> 73.81 | No clear effect |
| set similarity loss weight to 0.1 instead of 0.05 (ablation) | ERA-Net · NTU60 | AUC: 72.92 -> 73.87 | Improved |
Closest papers by meaning
- Open-Vocabulary Long Term Action Anticipation (ECCV 2026 · similarity 0.86)
- EAST: Early Action Prediction Sampling Strategy with Token Masking (ICLR 2026 · similarity 0.86)
- Action Sequence Augmentation for Action Anticipation (ICLR 2025 · similarity 0.85)
- Progressive Teacher-Student Learning for Early Action Prediction (CVPR 2019 · similarity 0.85)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.83)
- On Discriminative vs. Generative classifiers: Rethinking MLLMs for Action Understanding (ICLR 2026 · similarity 0.83)
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.
