Research map / Vision-language and multimodal / Cross-modal learning
Modality Equilibrium Matters: Minor-Modality-Aware Adaptive Alternating for Cross-Modal Memory Enhancement
Xiang Shi, Rui Zhang, Yu Liu, Yinpeng Liu, Zhu Liang, Qikai Cheng, Wei Ping Lu
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 cross-modal alignment (ablation) | Ours · CREMA-D | accuracy: 84.41 -> 84.95 | Improved |
| add memory unit (ablation) | Ours · CREMA-D | accuracy: 64.25 -> 84.95 | Improved |
| add memory inheritance (ablation) | Ours · CREMA-D | accuracy: 78.36 -> 84.95 | Improved |
| add text modality to audio (ablation) | Ours · IEMOCAP | accuracy: 72.18 -> 75.07 | Improved |
| add visual modality to audio and text (ablation) | Ours · IEMOCAP | accuracy: 72.18 -> 75.07 | Improved |
| replace fixed threshold with learnable gate (ablation) | Ours · CREMA-D | accuracy: 84.95 -> 84.81 | No clear effect |
| replace fixed threshold with learnable gate (ablation) | Ours · Kinetics-400 | accuracy: 79.35 -> 80.21 | Improved |
| weak-to-strong order instead of strong-to-weak or random (ablation) | Ours · four benchmarks | accuracy | Improved |
| vary initial modality order (ablation) | Ours · four benchmarks | accuracy: less than 1% fluctuation | No clear effect |
| alternating EDM-guided training vs MLA | Ours vs MLA · Kinetics-400 | accuracy: 72.66 -> 79.35 | Improved |
| EDM-guided method vs MLA under missing modality | Ours vs MLA · IEMOCAP 10% missing | accuracy: 52.57 -> 69.56 | Improved |
| alternating training vs joint concat | Ours vs Concat · CREMA-D | accuracy: 70.30 -> 84.95 | Improved |
Closest papers by meaning
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 2025 · similarity 0.89)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.89)
- A Closer Look at Multimodal Representation Collapse (ICML 2025 · similarity 0.88)
- ReconBoost: Boosting Can Achieve Modality Reconcilement (ICML 2024 · similarity 0.88)
- Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs (NeurIPS 2026 · similarity 0.87)
- Mitigating Intra- and Inter-modal Forgetting in Continual Learning of Unified Multimodal Models (NeurIPS 2025 · similarity 0.87)
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.
