Research map / Large language models / Retrieval and knowledge
Self-Correction Distillation for Structured Data Question Answering
Yushan Zhu, Wen Zhang, Long Jin, Mengshu Sun, Ling Zhong, Zhiqiang Liu, Juan Li, Lei Liang, Chong Long, Chao Deng, Junlan Feng
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 self-distillation stage after teacher distillation (ablation) | SCD Llama3.1-8B · WTQ | Denotation accuracy: 46.2 -> 48.9 | Improved |
| add self-distillation stage (ablation) | SCD Llama3.1-8B · CronQuestion Complex | Hit@1: 93.6 -> 95.1 | Improved |
| provide detailed error messages (type-specific) in EPM (ablation) | SCD Llama3.1-8B · CronQuestion Complex | Hit@1: 90.9 -> 95.1 | Improved |
| provide detailed error messages in EPM (ablation) | SCD Llama3.1-8B · WebQSP | Hit@1: 79.5 -> 83.7 | Improved |
| add EPM at inference for fine-tuned student | SCD Llama3.1-8B · WTQ | Denotation accuracy: 43.9 -> 48.9 | Improved |
| add EPM at inference for GPT4 teacher | GPT4 · WebQSP | Hit@1: 84.4 -> 91.7 | Improved |
| add EPM at inference for untuned Llama3.1 | Llama3.1-8B · WikiSQL | Denotation accuracy: 73.4 -> 74.5 | No clear effect |
| SCD vs KPOD distillation | Llama3.1-8B · CronQuestion Complex | Hit@1: 92.4 -> 95.1 | Improved |
| SCD vs PERsD distillation | Llama3.1-8B · WTQ | Denotation accuracy: 44.3 -> 48.9 | Improved |
| SCD vs KPOD on unseen TabFact | Llama3.1-8B · TabFact | Accuracy: 79.2 -> 83.4 | Improved |
| increase max correction time and demonstrations (ablation) | Llama3.1 and Naive-SFT · WebQSP | Hit@1 | No clear effect |
| increase max correction time and demonstrations (ablation) | SCD, KPOD, PERsD, GPT4 · WebQSP | Hit@1 | Improved |
Closest papers by meaning
- TrustUQA: A Trustful Framework for Unified Structured Data Question Answering (AAAI 2025 · similarity 0.88)
- Diagnosing and Addressing Pitfalls in KG-RAG Datasets: Toward More Reliable Benchmarking (NeurIPS 2025 · similarity 0.86)
- FlexKBQA: A Flexible LLM-Powered Framework for Few-Shot Knowledge Base Question Answering (AAAI 2024 · similarity 0.86)
- CITE: Anytime Valid Statistical Inference in LLM Self-Consistency (NeurIPS 2026 · similarity 0.86)
- DDK: Distilling Domain Knowledge for Efficient Large Language Models (NeurIPS 2024 · similarity 0.86)
- SuRe: Summarizing Retrievals using Answer Candidates for Open-domain QA of LLMs (ICLR 2024 · similarity 0.86)
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.
