Research map / Vision-language and multimodal / Human and embodied modeling
Learning to Scale Multilingual Representations for Vision-Language Tasks
Andrea Burns, Donghyun Kim, Derry Wijaya, Kate Saenko, Bryan A. Plummer
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 language-specific embeddings to the language-agnostic vocabulary (hybrid embedding model) (ablation) | Language-Agnostic · MSCOCO multilingual retrieval | average mR (A): 73.5 -> 76.9 | Improved |
| add language-specific embeddings to the language-agnostic vocabulary (hybrid embedding model) (ablation) | Language-Agnostic · Multi30K multilingual retrieval | average mR (A): 61.6 -> 64.0 | Improved |
| add Masked Cross-Language Modeling loss (ablation) | HEM · MSCOCO multilingual retrieval | average mR (A): 76.9 -> 77.7 | Improved |
| add Masked Cross-Language Modeling loss (ablation) | HEM · Multi30K multilingual retrieval | average mR (A): 64.0 -> 65.4 | Improved |
| add CLC-A (average ensemble over translated queries) (ablation) | SMALR · MSCOCO multilingual retrieval | human-average mR (HA): 77.7 -> 78.6 | Improved |
| add CLC-A (average ensemble over translated queries) (ablation) | SMALR · Multi30K multilingual retrieval | human-average mR (HA): 65.0 -> 68.2 | Improved |
| replace CLC-A with weighted CLC-C classifier ensemble (ablation) | SMALR-CLC-A · MSCOCO multilingual retrieval | human-average mR (HA): 78.6 -> 79.7 | Improved |
| replace CLC-A with weighted CLC-C classifier ensemble (ablation) | SMALR-CLC-A · Multi30K multilingual retrieval | human-average mR (HA): 68.2 -> 68.7 | Improved |
| add exploration parameters (p=0.2, M=20) when mapping to the latent vocabulary (ablation) | Language-Agnostic · Multi30K multilingual retrieval | average mR (A): 72.1 -> 73.5 | Improved |
| add exploration parameters (p=0.2, M=20) when mapping to the latent vocabulary (ablation) | Language-Agnostic · MSCOCO multilingual retrieval | average mR (A): 60.0 -> 61.6 | Improved |
| vary the weight lambda_2 of the masking loss (ablation) | SMALR · Multi30K multilingual retrieval | average mR | No clear effect |
| use SMALR instead of S-LIWE as the multilingual representation | S-LIWE (10) · MSCOCO multilingual retrieval | average mR (A): 66.7 -> 77.7 | Improved |
Closest papers by meaning
- LRM-LLaVA: Overcoming the Modality Gap of Multilingual Large Language-Vision Model for Low-Resource Languages (AAAI 2025 · similarity 0.90)
- Embracing Language Inclusivity and Diversity in CLIP through Continual Language Learning (AAAI 2024 · similarity 0.89)
- DiM$^3$: Bridging Multilingual and Multimodal Models via Direction- and Magnitude-Aware Merging (NeurIPS 2026 · similarity 0.89)
- Towards General Continuous Memory for Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Multilingual Diversity Improves Vision-Language Representations (NeurIPS 2024 · similarity 0.88)
- MMRL: Multi-Modal Representation Learning for Vision-Language Models (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.
