Research map / Vision-language and multimodal / CLIP and zero-shot
Universal Multimodal Representation for Language Understanding
Zhuosheng Zhang, Kehai Chen, Rui Wang, Masao Utiyama, Eiichiro Sumita, Zuchao Li, Hai Zhao
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 UVR-TILT image retrieval | Transformer-Base · WMT16 En-Ro | BLEU: 32.66 -> 33.78 | Improved |
| add UVR-CMRM image retrieval (COCO) | Transformer-Big · WMT14 En-De | BLEU: 28.45 -> 29.21 | Improved |
| add UVR-TILT to BERT-WWM | BERT-WWM · GLUE RTE test | accuracy: 77.3 -> 84.5 | Improved |
| add UVR-TILT to BERT-WWM | BERT-WWM · CoLA | MCC: 63.6 -> 62.5 | Hurt |
| add UVR-TILT to MT-DNN large | MT-DNN large · GLUE RTE validation | accuracy: 78.70 -> 82.19 | Improved |
| add UVR-TILT to MT-DNN base | MT-DNN base · GLUE QNLI validation | accuracy: 90.32 -> 91.31 | Improved |
| shuffle image features but keep lookup table (ablation) | UVR-TILT · WMT16 En-Ro | BLEU: 33.72 -> 33.40 | Improved |
| keep trained image features instead of random init (ablation) | UVR-TILT · WMT16 En-Ro | BLEU: 33.08 -> 33.72 | Improved |
| keep sentence-image mapping instead of random mapping (ablation) | UVR-TILT · WMT16 En-Ro | BLEU: 32.05 -> 33.72 | Improved |
| use ResNet152 instead of ResNet50 features (ablation) | UVR-TILT · WMT16 En-Ro | BLEU: 33.72 -> 33.82 | No clear effect |
| add UVR-TILT when source content words are removed | Transformer · Multi30K En-De test2016 | BLEU: 10.94 -> 12.80 | Improved |
| add UVR-CMRM | Transformer · WAT'19 En-Hi challenge | BLEU: 20.52 -> 21.62 | Improved |
Closest papers by meaning
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.89)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.89)
- MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer (ICLR 2026 · similarity 0.89)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.88)
- Chat-UniVi: Unified Visual Representation Empowers Large Language Models with Image and Video Understanding (CVPR 2024 · 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.
