Research map / Vision-language and multimodal / Human and embodied modeling
MULE: Multimodal Universal Language Embedding
Donghyun Kim, Kuniaki Saito, Kate Saenko, Stan Sclaroff, Bryan Plummer
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 a shared, visually-semantic multilingual embedding rather than separate language branches | Image-sentence retrieval · Multi30K French | mean recall: 52.0 -> 56.4 | Improved |
| Use a shared, visually-semantic multilingual embedding rather than separate language branches | Image-sentence retrieval · Multi30K Czech | mean recall: 43.2 -> 50.3 | Improved |
| Use a shared multilingual model instead of an English-only model | Image-sentence retrieval · Multi30K English | mean recall: 71.1 -> 68.0 | Hurt |
| Add multi-layer neighborhood constraints (ablation) | MULE · Multi30K Czech | mean recall: 43.9 -> 46.4 | Improved |
| Add a language classifier to neighborhood constraints (ablation) | MULE · Multi30K Czech | mean recall: 46.4 -> 49.5 | Improved |
| Add MULE pretraining (ablation) | MULE with neighborhood constraints and language classifier · Multi30K Czech | mean recall: 49.5 -> 50.3 | Improved |
| Add neighborhood constraints (ablation) | MULE · MSCOCO Chinese | mean recall: 54.2 -> 56.6 | Improved |
| Add a language classifier to neighborhood constraints (ablation) | MULE · MSCOCO Chinese | mean recall: 56.6 -> 57.9 | Improved |
| Add MULE pretraining (ablation) | MULE with neighborhood constraints and language classifier · MSCOCO Chinese | mean recall: 57.9 -> 58.8 | Improved |
| Add machine-translated training sentences alongside human-generated sentences | MULE EmbN Full · Multi30K French | mean recall: 56.4 -> 62.3 | Improved |
| Add machine-translated training sentences alongside human-generated sentences | MULE EmbN Full · Multi30K Czech | mean recall: 50.3 -> 57.7 | Improved |
| Add machine-translated training sentences alongside human-generated sentences | MULE EmbN Full · MSCOCO Chinese | mean recall: 58.8 -> 74.8 | Improved |
Closest papers by meaning
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.90)
- Multilingual Diversity Improves Vision-Language Representations (NeurIPS 2024 · similarity 0.89)
- Towards General Continuous Memory for Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- Embracing Language Inclusivity and Diversity in CLIP through Continual Language Learning (AAAI 2024 · similarity 0.88)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · similarity 0.88)
- Bridging Vision and Language Spaces with Assignment Prediction (ICLR 2024 · 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.
