arXiv:2606.02607cs.LGcs.AI2026-06中稿 · ICML

用几何关系显式监督表格生成,提升隐私保护数据合成效果。

Geometry-Aware Tabular Diffusion

论文配图:Geometry-Aware Tabular Diffusion
图 1 · 摘自论文原文
  • 通过列间值差计算角度和长度,显式建模列关系。
  • 在10个数据集上9项指标领先,参数量减少3.5倍(分类任务达25倍)。
  • 方法可迁移至GNN与Transformer,适合数据合成与隐私保护研究者。

表格数据合成对隐私保护共享与数据增强至关重要,但现有扩散模型依赖隐式机制捕捉列间关系。我们提出几何感知表格扩散(GATD),将列值差产生的成对角度和长度作为输入与辅助目标,注入扩散去噪器。其基于MLP的实现,在十项基准测试中取得最优表现:平均参数量减少3.5倍(分类任务最多减少25倍),在10个数据集上8/10胜出形状一致性,7/10胜出趋势一致性,9/10在下游任务中(F1/RMSE)表现更优,形状与趋势误差分别降低27%和20%。默认损失权重可迁移至GNN与Transformer去噪器,在27/30和25/30的架构-数据集组合中提升形状与趋势表现。配对消融实验表明,性能提升源于关系监督(非额外输入或容量)。这表明显式关系监督是表格扩散中的通用归纳偏置。

原文摘要 · Abstract (English)

Tabular synthesis is critical for privacy-preserving sharing and augmentation, yet diffusion models rely on implicit mechanisms to capture inter-column relationships. We introduce Geometry-Aware Tabular Diffusion (GATD), which augments tabular diffusion denoisers with pairwise angles and lengths computed from column value differences and used as inputs and auxiliary targets. Our MLP instantiation achieves state-of-the-art benchmark performance while using 3.5x fewer parameters on average (up to 25x for classification tasks): on ten datasets, it wins 8/10 Shape, 7/10 Trend, and 9/10 downstream utility (F1/RMSE), reducing Shape and Trend error by 27% and 20%. Default loss weights transfer to GNN and Transformer denoisers, improving Shape on 27/30 and Trend on 25/30 architecture-dataset cells. A matched ablation shows supervision (not extra inputs or capacity) drives the gain. This shows explicit relational supervision is a portable inductive bias for tabular diffusion.

表格生成扩散模型隐私保护几何先验

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。