Research map / Vision-language and multimodal / Human and embodied modeling
MAP: Multimodal Uncertainty-Aware Vision-Language Pre-Training Model
Yatai Ji, Junjie Wang, Yuan Gong, Lin Zhang, Yanru Zhu, Hongfa Wang, Jiaxing Zhang, Tetsuya Sakai, Yujiu Yang
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 |
|---|---|---|---|
| Represent modalities as probability distributions with PDE rather than point features (ablation) | MAP · VQA2.0 test-dev, MSCOCO pretraining | accuracy: 74.57 -> 75.01 | Improved |
| Represent modalities as probability distributions with PDE rather than point features (ablation) | MAP · SNLI-VE val, MSCOCO pretraining | accuracy: 79.42 -> 80.05 | Improved |
| Represent modalities as probability distributions with PDE rather than point features (ablation) | MAP · NLVR2 dev, MSCOCO pretraining | accuracy: 77.72 -> 78.96 | Improved |
| Add sequence-level interaction to PDE (ablation) | MAP · VQA2.0 test-dev | accuracy: 72.01 -> 73.35 | Improved |
| Use Softmax rather than Sigmoid+Normal in PDE (ablation) | MAP · VQA2.0 test-dev | accuracy: 73.34 -> 73.35 | Improved |
| Use ReLU+Normal rather than Softmax in PDE (ablation) | MAP · VQA2.0 test-dev | accuracy: 73.35 -> 69.70 | Hurt |
| Use ReLU²+Normal rather than Softmax in PDE (ablation) | MAP · VQA2.0 test-dev | accuracy: 73.35 -> 70.53 | Hurt |
| Pretrain with D-MLM, D-ITM, and D-VLC instead of random initialization (ablation) | MAP · VQA2.0 test-dev | accuracy: 73.35 -> 75.16 | Improved |
| Include D-MLM in distribution-based pretraining (ablation) | MAP · SNLI-VE test-p | accuracy: 78.54 -> 80.39 | Improved |
| Use D-MLM and D-ITM without D-VLC rather than all three objectives (ablation) | MAP · NLVR2 test | accuracy: 79.47 -> 79.64 | Improved |
| Increase cross-modal transformer depth from 6 to 8 layers after pretraining (ablation) | MAP · VQA2.0 test-dev | accuracy: 75.16 -> 75.26 | Improved |
| Increase cross-modal transformer depth from 6 to 8 layers with random initialization (ablation) | MAP · VQA2.0 test-dev | accuracy: 73.35 -> 73.31 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VLMo-Base — VLMo: Unified Vision-Language Pre-Training with Mixture-of-Modality-Experts (NeurIPS 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FUSE: Quantifying Uncertainty in Vision-Language Models by Bayesian Fusing Epistemic and Aleatoric Uncertainty (ICML 2026 · similarity 0.90)
- Exploiting the Asymmetric Uncertainty Structure of Pre-trained VLMs on the Unit Hypersphere (NeurIPS 2025 · similarity 0.89)
- Probabilistic Language-Image Pre-Training (ICLR 2025 · similarity 0.89)
- CertainlyUncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awareness (ICLR 2025 · similarity 0.88)
- LION: Empowering Multimodal Large Language Model with Dual-Level Visual Knowledge (CVPR 2024 · similarity 0.87)
- Unified Generative and Discriminative Training for Multi-modal Large Language Models (NeurIPS 2024 · similarity 0.87)
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.
