Research map / Vision-language and multimodal / Visual question answering
Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 an explanation to the answer generation | UnifiedQA-BASE fine-tuning · ScienceQA test | accuracy: 70.12 -> 73.33 (+3.21%) | Helped |
| Add a lecture and explanation to answer generation | UnifiedQA-BASE fine-tuning · ScienceQA test | accuracy: 70.12 -> 74.11 (+3.99%) | Helped |
| Add an explanation to the answer generation | GPT-3 2-shot prompting · ScienceQA test | accuracy: 73.97 -> 74.61 | Mixed |
| Add a lecture and explanation to answer generation | GPT-3 2-shot prompting · ScienceQA test | accuracy: 73.97 -> 75.17 (+1.20%) | Helped |
| Provide the gold lecture and explanation in the test prompt instead of having the model generate them | GPT-3 · ScienceQA test | accuracy: 18.96% | Helped |
| Provide the gold explanation and lecture in the test prompt | GPT-3 · ScienceQA test | accuracy: 18.86% | Helped |
| Generate the lecture and explanation before the answer (ablation) | GPT-3 (CoT) · ScienceQA test, 1,000 test examples | accuracy | Hurt |
| Use two in-context examples rather than zero-shot prompting (ablation) | GPT-3 · ScienceQA test | accuracy: 74.04 -> 73.97 | No clear effect |
| Select in-context examples dynamically from the same topic, category, or skill (ablation) | GPT-3 (CoT, ALE) · ScienceQA test | accuracy: 75.15; 74.58; 75.10 | No clear effect |
| Add lectures as well as explanations to the GPT-3 CoT prompt (ablation) | GPT-3 (CoT) · ScienceQA test | accuracy: QCM→ALE has the highest accuracy on average | Helped |
| Use only explanations rather than lectures and explanations in the GPT-3 CoT prompt (ablation) | GPT-3 (CoT) · ScienceQA test | accuracy: 74.61 vs. 75.17 | Hurt |
| Increase the number of in-context examples beyond two (ablation) | GPT-3 (CoT, ALE) · ScienceQA test | accuracy | Hurt |
Closest papers by meaning
- T-SciQ: Teaching Multimodal Chain-of-Thought Reasoning via Large Language Model Signals for Science Question Answering (AAAI 2024 · similarity 0.94)
- KAM-CoT: Knowledge Augmented Multimodal Chain-of-Thoughts Reasoning (AAAI 2024 · similarity 0.90)
- MicroVQA: A Multimodal Reasoning Benchmark for Microscopy-Based Scientific Research (CVPR 2025 · similarity 0.89)
- MME-CoT: Benchmarking Chain-of-Thought in Large Multimodal Models for Reasoning Quality, Robustness, and Efficiency (ICML 2025 · similarity 0.89)
- SPIQA: A Dataset for Multimodal Question Answering on Scientific Papers (NeurIPS 2024 · similarity 0.89)
- MME-SCI: A Comprehensive and Challenging Science Benchmark for Multimodal Large Language Models (AAAI 2026 · 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.