AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

ReTabSyn: Realistic Tabular Data Synthesis via Reinforcement Learning

Xiaofeng Lin, Seungbae Kim, Zhuoya Li, Zachary DeSoto, Charles Fleming, Guang Cheng

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias · 0 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
fine-tune GReaT generator with DPO on oracle-free perturbation pairsGReaT (GPT-2) · Adult 1% prevalencePR-AUC: 0.873 -> 0.906Improved
add DPO fine-tuning after augmented pre-training (ablation)ReTabSyn · WiltAUROC: 0.815 -> 0.841Improved
add augmented pre-training before DPO (ablation)ReTabSyn · WiltAUROC: 0.800 -> 0.841Improved
use oracle-free perturbation instead of classifier-based editing (ablation)ReTabSyn · WiltAUROC: 0.809 -> 0.841Improved
use oracle-free perturbation instead of classifier-based screening (ablation)ReTabSyn · WiltAUROC: 0.817 -> 0.841Improved
use only target-perturbation pairs (proportion 1) rather than 50/50 mix (ablation)ReTabSyn · WiltAUROC / Corr.: AUROC 0.841 -> 0.843; Corr. 0.880 -> 0.862Mixed
use only feature-perturbation pairs (proportion 0) (ablation)ReTabSyn · WiltAUROC / Corr.: AUROC 0.841 -> 0.821; Corr. 0.880 -> 0.888Mixed
use dynamic augmentation size instead of fixed 1000 rows (ablation)ReTabSyn · WiltAUROC: 0.835 -> 0.841Improved
increase augmentation to Nx20 (ablation)ReTabSyn · WiltAUROC: 0.841 -> 0.837No clear effect
add DPO-Positive regularizer (lambda 0.1 vs 0) (ablation)ReTabSyn · WiltAUROC: 0.772 -> 0.841Improved
vary DPO beta over 0.05, 0.1, 0.2 (ablation)ReTabSyn · WiltAUROC: 0.837 / 0.841 / 0.839No clear effect
apply ReTabSyn alignment to Qwen3.5-2B backboneQwen3.5-2B · Wilt (synthetic only)AUROC: 0.870 -> 0.900Improved

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