AI Paper Atlas

Research map / Large language models / Agents, code and math

Better Datasets Start from RefineLab: Automatic Optimization for High-Quality Dataset Refinement

Xiaonan Luo, Yue Huang, Ping He, Xiangliang Zhang

AAAI 2026 · Large language models › Agents, code and math · approach: LLM-guided program synthesis · 2 citations (OpenAlex)

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 changedWhereMeasuredEffect
use ILP assignment instead of greedy at 0.25CRefineLab · MMLU 10% pilotcoverage JSD: 0.252 -> 0.183Improved
use ILP assignment instead of greedy at 0.25CRefineLab · MMLU 10% pilotdifficulty JSD: 0.328 -> 0.388Hurt
use ILP assignment instead of greedy at 0.5CRefineLab · MMLU 10% pilotcoverage JSD: 0.190 -> 0.063Improved
use ILP assignment instead of greedy at 1.0CRefineLab · MMLU 10% pilotdistractor diversity: 0.300 -> 0.331Improved
use ILP assignment instead of uniform budget split at 1.0CRefineLab · MMLU 10% pilotdifficulty JSD: 0.412 -> 0.295Improved
increase budget from 0.25C to 1.0CRefineLab · MMLU 10% pilotcoverage JSD: 0.183 -> 0.019Improved
refine GSM8K with RefineLabGSM8K · GSM8Kperformance spread of top models: 2.2% -> 11.8%Improved
refine GSM8K with RefineLabGSM8K · GSM8Kgap o4-mini vs Mistral-8x7B: 30.8% -> 49.4%Improved
refine HumanEval with RefineLabHumanEval · HumanEvalperformance spread of top models: 0.6% -> 13.4%Improved
refine GSM8K with RefineLabGSM8K · GSM8K prompting variantsaverage performance range across LLMs: 8.4% -> 4.9%Improved
fine-tune on RefineLab-refined data instead of originalLlama-3.1-8B · GSM8K 150-item testaccuracy: 58.7% -> 65.6%Improved
fine-tune on RefineLab-refined data instead of originalQwen-2-7B · GSM8K 150-item testaccuracy: 67.3% -> 72.7%Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy