Research map / Vision-language and multimodal / Visual question answering
Collaborative Enhancement of Large and Small Models for Question Answering via Dual Knowledge Transfer
Shaofei Wang, Yunan Liu, Xiaolan Tang, Wenlong Chen
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 confidence-aware prompting (ablation) | DKT large model (DeepSeek r1-14B) · ReClor | accuracy: 71.8 -> 73.6 | Improved |
| add distillation substep (ablation) | DKT small model · ReClor | accuracy: 57.6 -> 59.6 | Improved |
| add judgment and reflection substeps (ablation) | DKT small model · ReClor | accuracy: 57.8 -> 59.6 | Improved |
| add iteration between small4large and large4small (ablation) | DKT small model · ReClor | accuracy: 59.0 -> 59.6 | Improved |
| add iteration between small4large and large4small (ablation) | DKT large model · ReClor | accuracy: 72.8 -> 73.6 | Improved |
| add distillation substep (ablation) | DKT small model · LogiQA | accuracy: 34.25 -> 37.33 | Improved |
| add confidence-aware prompting (ablation) | DKT large model (DeepSeek r1-14B) · LogiQA | accuracy: 62.98 -> 64.67 | Improved |
| apply DKT to the small model with LLaMA 3-70B as large model | RoBERTa-based small model · CSQA | accuracy: 60.69 -> 63.96 | Improved |
| apply DKT to the small model with DeepSeek r1-14B as large model, even when large model is worse | small model · ProtoQA | Max Answers@10: 49.1 -> 51.7 | Improved |
| apply DKT large4small and small4large to LLaMA 3-70B | LLaMA 3-70B · ReClor | accuracy: 62.6 -> 66.0 | Improved |
| apply DKT to LLaMA 3-70B | LLaMA 3-70B · ProtoQA | Max Answers@10: 38.2 -> 40.2 | No clear effect |
| run more than two iterations of transfer (ablation) | DKT · LogiQA and ReClor | accuracy | No clear effect |
Closest papers by meaning
- Rethinking Generative Large Language Model Evaluation for Semantic Comprehension (ICML 2024 · similarity 0.89)
- Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation (AAAI 2025 · similarity 0.88)
- Cost-efficient Knowledge-based Question Answering with Large Language Models (NeurIPS 2024 · similarity 0.88)
- BLADE: Enhancing Black-Box Large Language Models with Small Domain-Specific Models (AAAI 2025 · similarity 0.88)
- FlexKBQA: A Flexible LLM-Powered Framework for Few-Shot Knowledge Base Question Answering (AAAI 2024 · similarity 0.88)
- SuRe: Summarizing Retrievals using Answer Candidates for Open-domain QA of LLMs (ICLR 2024 · 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.
