Research map / Vision-language and multimodal / Multimodal LLMs
Bootstrapping Grounded Chain-of-Thought in Multimodal LLMs for Data-Efficient Model Adaptation
Jiaer Xia, Bingkui Tong, Yuhang Zang, Rui Shao, Kaiyang Zhou
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 box verification of CoT (ablation) | GCoT · TabMWP (8 samples) | accuracy: 17.92 -> 23.86 | Improved |
| add box verification of CoT (ablation) | GCoT · TabMWP (128 samples) | accuracy: 23.57 -> 33.93 | Improved |
| add augmentation of three verified GCoTs per question (ablation) | GCoT · TabMWP (8 samples) | accuracy: 21.56 -> 23.86 | Improved |
| add augmentation of three verified GCoTs per question (ablation) | GCoT · TabMWP (128 samples) | accuracy: 31.57 -> 33.93 | Improved |
| add box grounding to distilled CoT (LLaMA3.2 source) | Distillation · TabMWP (32 samples) | accuracy: 24.14 -> 27.35 | Improved |
| add box grounding to distilled CoT (Claude3.5 source) | Distillation · TabMWP (32 samples) | accuracy: 19.84 -> 25.46 | Improved |
| add box grounding to distilled CoT (GPT-4o source) | Distillation · TabMWP (32 samples) | accuracy: 25.91 -> 27.46 | Improved |
| add box grounding, averaged over five CoT sources | Distillation · TabMWP (32 samples) | accuracy: 22.78 -> 26.54 | Improved |
| train on distilled CoT instead of question-answer pairs | Fine-tuning · ChartQA, 64 to 128 samples | accuracy | Mixed |
| add more bootstrapping loop iterations (ablation) | GCoT bootstrapping · five datasets | number of true boxes | Improved |
| replace CoT with GCoT when trained on 128 ChartQA samples | CoT · TabMWP transfer | accuracy: 21.01 -> 25.55 | Improved |
| replace CoT with GCoT when trained on 128 ChartQA samples | CoT · DVQA transfer | accuracy: 12.34 -> 15.73 | Improved |
Closest papers by meaning
- TVI-CoT: Text-Visual Interleaved Chain-of-Thought Reasoning for Multimodal Understanding (ICML 2026 · similarity 0.90)
- Uni-CoT: Towards Unified Chain-of-Thought Reasoning Across Text and Vision (ICLR 2026 · similarity 0.89)
- Compositional Chain-of-Thought Prompting for Large Multimodal Models (CVPR 2024 · similarity 0.89)
- Thinking Diffusion: Penalize and Guide Visual-Grounded Reasoning in Diffusion Multimodal Language Models (CVPR 2026 · similarity 0.89)
- L2V-CoT: Cross-Modal Transfer of Chain-of-Thought Reasoning via Latent Intervention (AAAI 2026 · similarity 0.89)
- Grounded Chain-of-Thought for Multimodal Large Language Models (CVPR 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.
