Research map / Vision-language and multimodal / Cross-modal learning
Multimodal Distribution Matching for Vision-Language Dataset Distillation
Jongoh Jeong, Hoyong Kwon, Minseok Kim, Kuk-Jin Yoon
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 |
|---|---|---|---|
| initialize synthetic data with joint-feature k-means instead of random real samples (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.6 -> 21.9 | Improved |
| initialize synthetic data with joint k-means instead of Gaussian noise (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 0.5 -> 21.9 | Improved |
| cluster on joint image-text features instead of image-only features (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.8 -> 21.9 | Improved |
| use joint k-means seeding instead of K-center seeding (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.8 -> 21.9 | Improved |
| use joint k-means seeding instead of herding seeding (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 21.5 -> 21.9 | Improved |
| use fine-tuned expert instead of pre-trained model as frozen model (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 6.8 -> 17.8 | Improved |
| use proposed angle-based multi-expert interpolation instead of fine-tuned model alone (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 17.8 -> 21.9 | Improved |
| use proposed interpolation instead of random weighted sum of pre-trained and fine-tuned (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 19.6 -> 21.9 | Improved |
| add agreement loss to InfoNCE (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.98 -> 21.02 | No clear effect |
| add discrepancy loss to InfoNCE (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.98 -> 21.53 | Improved |
| add both agreement and discrepancy losses to InfoNCE (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.98 -> 21.94 | Improved |
| use geodesic kernel instead of Laplacian RBF kernel (ablation) | MDM · Flickr8k (100 pairs) | mean retrieval recall: 20.3 -> 21.9 | Improved |
Closest papers by meaning
- CovMatch: Cross-Covariance Guided Multimodal Dataset Distillation with Trainable Text Encoder (NeurIPS 2025 · similarity 0.91)
- Topology-Aware Representation Alignment for Semi-Supervised Vision-Language Learning (NeurIPS 2026 · similarity 0.89)
- Efficient Multimodal Dataset Distillation via Generative Models (NeurIPS 2025 · similarity 0.88)
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.88)
- Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation (NeurIPS 2025 · similarity 0.88)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · 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.
