Research map / Vision-language and multimodal / Cross-modal learning
Better Together: Leveraging Unpaired Multimodal Data for Stronger Unimodal Models
Sharut Gupta, Shobhita Sundaram, Chenyu Wang, Stefanie Jegelka, Phillip Isola
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 unpaired text with shared weights (UML) | self-supervised unimodal model · MUSTARD | accuracy: 59.66 -> 63.28 | Improved |
| add unpaired text with shared weights (UML) | self-supervised unimodal model · UR-FUNNY | accuracy: 56.99 -> 57.34 | No clear effect |
| add unpaired text (UML) | DINOv2 ViT-S/14 full fine-tuning · 9 vision datasets average | accuracy: 81.54 -> 83.99 | Improved |
| add unpaired text (UML) | DINOv2 ViT-S/14 full fine-tuning · SUN397 | accuracy: 66.20 -> 66.03 | Hurt |
| add unpaired text (UML) | DINOv2 linear probe · 9 vision datasets average, 1-shot | accuracy: 45.52 -> 51.36 | Improved |
| add unpaired text (UML) | DINOv2 linear probe · 9 vision datasets average, 4-shot | accuracy: 65.84 -> 68.53 | Improved |
| use CLIP aligned encoders instead of DINOv2 + OpenLLaMA (ablation) | UML · image classification | accuracy | Improved |
| train with semantically unrelated modality (ablation) | UML · image classification | accuracy | No clear effect |
| unfreeze the text encoder (ablation) | UML · image classification | accuracy | Mixed |
| initialize ViT transformer with pretrained BERT weights | ViT · ImageNet | accuracy | Improved |
| add audio, vision and text auxiliary modalities (ablation) | UML · ImageNet-ESC audio classification | accuracy | Improved |
Closest papers by meaning
- MSP: Modality Self-Play for Training Multimodal LLMs Without Paired Cross-Modal Data (NeurIPS 2026 · similarity 0.90)
- Learning Shared Representations from Unpaired Data (NeurIPS 2025 · similarity 0.90)
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model (CVPR 2025 · similarity 0.88)
- TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs (NeurIPS 2026 · similarity 0.88)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.88)
- LoMo: Local Modality Substitution for Deeper Vision-Language Fusion (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.
