Research map / Vision-language and multimodal / Visual question answering
A Deep Reinforcement Learning Based Multi-Step Coarse to Fine Question Answering (MSCQA) System
Yu Wang, Hongxia Jin
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 action a3 (reject false-positive answers) and sub-context selection (ablation) | MSCQA · WikiReading | answer accuracy (%): 74.3 -> 75.8 | Improved |
| add action a3 (reject false-positive answers) and sub-context selection (ablation) | MSCQA · WR-Long | answer accuracy (%): 42.8 -> 43.6 | Improved |
| add action a3 (reject false-positive answers) and sub-context selection (ablation) | MSCQA · CNN | answer accuracy (%): 76.3 -> 77.6 | Improved |
| add action a3 (reject false-positive answers) and sub-context selection (ablation) | MSCQA · SQuAD1.1 dev | answer accuracy (%): 80.5 -> 83.6 | Improved |
| multi-step action selection instead of fixed single pipeline (Base Model 1) | MSCQA · WikiReading | training speed (samples/sec): 112 -> 186 | Improved |
| multi-step action selection instead of fixed single pipeline (Base Model 1) | MSCQA · WR-Long | training speed (samples/sec): 66 -> 129 | Improved |
| multi-step action selection instead of fixed single pipeline (Base Model 1) | MSCQA · CNN | training speed (samples/sec): 83 -> 173 | Improved |
| multi-step action selection instead of fixed single pipeline (Base Model 1) | MSCQA · SQuAD | training speed (samples/sec): 146 -> 213 | Improved |
| MSCQA vs QANet-style Base Model 2 on long documents | MSCQA · WR-Long | answer accuracy (%): 39.5 -> 43.6 | Improved |
| MSCQA vs coarse-to-fine Base Model 1 on short documents | MSCQA · SQuAD1.1 dev | answer accuracy (%): 75.2 -> 83.6 | Improved |
| MSCQA vs strong baseline on CNN | MSCQA · CNN | answer accuracy (%): 75.8 -> 77.6 | Improved |
| MSCQA vs Oracle sentence selection | MSCQA · SQuAD1.1 dev | answer accuracy (%): 86.2 vs 83.6 | Hurt |
Closest papers by meaning
- FrugalRAG: Less is More in RL Finetuning for Multi-hop Question Answering (ICLR 2026 · similarity 0.86)
- Ask the Right Questions: Active Question Reformulation with Reinforcement Learning (ICLR 2018 · similarity 0.86)
- Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning (ICLR 2025 · similarity 0.85)
- Cost-efficient Knowledge-based Question Answering with Large Language Models (NeurIPS 2024 · similarity 0.84)
- Thinker: Learning to Think Fast and Slow (NeurIPS 2025 · similarity 0.84)
- Putting People in LLMs’ Shoes: Generating Better Answers via Question Rewriter (AAAI 2025 · similarity 0.83)
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.
