Research map / Large language models / Reasoning and chain of thought
DDCoT: Duty-Distinct Chain-of-Thought Prompting for Multimodal Reasoning in Language Models
Ge Zheng, Bin Yang, Jiajin Tang, Hong-Yu Zhou, Sibei Yang
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 |
|---|---|---|---|
| Use DDCoT instead of GPT-3 (CoT) | GPT-3 on ScienceQA · ScienceQA test, IMG | accuracy: 67.43 -> 69.96 | Improved |
| Use DDCoT instead of Chameleon | ChatGPT on ScienceQA · ScienceQA test, IMG | accuracy: 70.80 -> 72.53 | Improved |
| Use DDCoT instead of MM-CoT | UnifiedQA on ScienceQA · ScienceQA test, Avg | accuracy: +2.43% | Improved |
| Add DDCoT rationales | Flamingo on ScienceQA · ScienceQA test, Avg | accuracy: 27.87 -> 39.01 | Improved |
| Add DDCoT rationales | MiniGPT-4 on ScienceQA · ScienceQA test, Avg | accuracy: 44.71 -> 55.67 | Improved |
| Add captions to the language-only baseline (ablation) | T5 baseline on ScienceQA · ScienceQA test, IMG | accuracy: 72.93 -> 75.26 | Improved |
| Use original image input without additional components (ablation) | T5 baseline on ScienceQA · ScienceQA test, IMG | accuracy: 72.93 -> 62.82 | Hurt |
| Add DDCoT rationales instead of using no rationale (ablation) | T5 baseline on ScienceQA · ScienceQA test, IMG | accuracy: 75.16 -> 83.34 | Improved |
| Replace naive rationale with decomposed-question rationale (ablation) | DDCoT rationale generation · ScienceQA test, IMG | accuracy: 75.06 -> 75.61 | Improved |
| Add duty-distinct prompting with negative-space uncertainty (ablation) | DDCoT rationale generation · ScienceQA test, IMG | accuracy: 75.61 -> 83.34 | Improved |
| Add explicit uncertainty prompting (ablation) | Duty-distinct rationale generation · ScienceQA test, IMG | accuracy: 78.19 -> 83.34 | Improved |
| Add joint reasoning integration (ablation) | DDCoT rationale generation · ScienceQA test, IMG | accuracy: 77.49 -> 83.34 | Improved |
Closest papers by meaning
- Uni-CoT: Towards Unified Chain-of-Thought Reasoning Across Text and Vision (ICLR 2026 · similarity 0.90)
- Grounded Chain-of-Thought for Multimodal Large Language Models (CVPR 2026 · similarity 0.90)
- TVI-CoT: Text-Visual Interleaved Chain-of-Thought Reasoning for Multimodal Understanding (ICML 2026 · similarity 0.90)
- MME-CoT: Benchmarking Chain-of-Thought in Large Multimodal Models for Reasoning Quality, Robustness, and Efficiency (ICML 2025 · similarity 0.89)
- Interleaved-Modal Chain-of-Thought (CVPR 2025 · similarity 0.89)
- KAM-CoT: Knowledge Augmented Multimodal Chain-of-Thoughts Reasoning (AAAI 2024 · similarity 0.89)
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.
