Research map / Vision-language and multimodal / Cross-modal learning
Stable Spectral Copula Alignment for Robust Multimodal Learning
Hongkang Zhang, Shao-Lun Huang, Yanlong Wang, Ercan Engin KURUOGLU
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 copula stabilization (ablation) | SSCA · CMU-MOSEI | degradation under 10x scaling: 4.8 -> 1.8 | Improved |
| add copula stabilization (ablation) | SSCA · MELD | degradation under 10x scaling: 4.5 -> 1.5 | Improved |
| add multiway hub coupling instead of pairwise (ablation) | SSCA · CMU-MOSEI | degradation under 10x scaling: 3.8 -> 1.8 | Improved |
| add multiway hub coupling instead of pairwise (ablation) | SSCA · MELD | degradation under 10x scaling: 3.5 -> 1.5 | Improved |
| add spectral alignment instead of contrastive loss (ablation) | SSCA · CMU-MOSEI | degradation under 10x scaling: 3.5 -> 1.8 | Improved |
| add spectral alignment instead of contrastive loss (ablation) | SSCA · MELD | degradation under 10x scaling: 3.2 -> 1.5 | Improved |
| add hub coupling and spectral stabilization to copula-only SigLIP | Copula-only + SigLIP · CMU-MOSEI | degradation under 10x scaling: 2.3 -> 1.8 | Improved |
| add hub coupling and spectral stabilization to copula-only SigLIP | Copula-only + SigLIP · MELD | degradation under 10x scaling: 2.2 -> 1.5 | Improved |
| add copula stabilization to SigLIP | SigLIP · CMU-MOSEI | degradation under 10x scaling: 6.3 -> 2.3 | Improved |
| use SSCA instead of MMML | MMML · CMU-MOSEI | degradation under 10x scaling: 3.2 -> 1.8 | Improved |
| use SSCA instead of CLIP baseline under JPEG Q=50 | CLIP ViT-B/16 · CC3M-500K Recall@1 | degradation under 10x scaling: 5.1 -> 2.5 | Improved |
| use SSCA instead of MMML on average real-world drift degradation | MMML · CMU-MOSEI and MELD drifts | degradation under 10x scaling: 2.2 -> 1.6 | Improved |
Closest papers by meaning
- Cross-Modal Alignment via Variational Copula Modelling (ICML 2025 · similarity 0.84)
- When to Align, When to Predict: A Phase Diagram for Multimodal Learning (NeurIPS 2026 · similarity 0.83)
- Propensity Score Alignment of Unpaired Multimodal Data (NeurIPS 2024 · similarity 0.82)
- PCA++: How Uniformity Induces Robustness to Background Noise in Contrastive Learning (NeurIPS 2025 · similarity 0.81)
- Bridging Multicalibration and Out-of-distribution Generalization Beyond Covariate Shift (NeurIPS 2024 · similarity 0.81)
- VCT: Training Consistency Models with Variational Noise Coupling (ICML 2025 · similarity 0.81)
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.
