Research map / Vision-language and multimodal / Cross-modal learning
Label What Matters: Modality-Balanced and Difficulty-Aware Multimodal Active Learning
Yuqiao Zeng, Xu Wang, Tengfei Liang, Yiqing Hao, Yi Jin, Hui Yu
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 |
|---|---|---|---|
| use RL-MBA instead of BMMAL (strongest baseline) | multimodal active learning · Food101 3,000 labels | Top-1 accuracy: 0.8609 -> 0.8650 | Improved |
| use RL-MBA instead of BMMAL | multimodal active learning · KineticsSound 3,000 labels | Top-1 accuracy: 0.4745 -> 0.4841 | Improved |
| use RL-MBA instead of BMMAL | multimodal active learning · VGGSound 3,000 labels | Top-1 accuracy: 0.2053 -> 0.2223 | Improved |
| use active learning strategies instead of random selection | baseline AL methods · VGGSound 3,000 labels | Top-1 accuracy: Random 0.2173 vs BMMAL 0.2053, BADGE 0.2023 | Hurt |
| add AMCB only to BMMAL baseline (ablation) | RL-MBA · Food101 / KineticsSound / VGGSound | Top-1 accuracy: 0.8609 -> 0.8621; 0.4745 -> 0.4771; 0.2053 -> 0.2059 | Improved |
| add EFDA only to BMMAL baseline (ablation) | RL-MBA · VGGSound | Top-1 accuracy: 0.2053 -> 0.2177 | Improved |
| combine AMCB and EFDA (full RL-MBA) (ablation) | RL-MBA · VGGSound | Top-1 accuracy: 0.2177 -> 0.2223 | Improved |
| use Relative reward instead of Absolute or Incremental reward (ablation) | RL-MBA policy · Food101 | Top-1 accuracy | Improved |
| use RL policy selection instead of heavy selection heuristics | RL-MBA · Food101 per-round time | round total time (s): 1151.39 -> 884.39 (vs BMMAL) | Improved |
| use RL-MBA instead of BMMAL on video-centric classes | KineticsSound class-wise · KineticsSound 5,000 labels | per-class accuracy | Mixed |
Closest papers by meaning
- Balancing Multimodal Training Through Game-Theoretic Regularization (NeurIPS 2025 · similarity 0.89)
- Enhancing Multimodal Cooperation via Sample-level Modality Valuation (CVPR 2024 · similarity 0.88)
- Revisiting the Data Sampling in Multimodal Post-training from a Difficulty-Distinguish View (AAAI 2026 · similarity 0.88)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.88)
- Facilitating Multimodal Classification via Dynamically Learning Modality Gap (NeurIPS 2024 · similarity 0.88)
- Fusion or Confusion? Multimodal Complexity Is Not All You Need (NeurIPS 2026 · 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.
