aiqbench analysis
This research exposes a critical blind spot in current synthetic data evaluation metrics, which often fail to detect the loss of inter-column dependencies essential for minority class performance. The study demonstrates that standard flow-matching objectives struggle with high-order correlations regardless of model capacity increases or common post-hoc fixes like copula correction. For practitioners relying on tabular generative models for fraud detection and clinical risk assessment, this implies a need to adopt new diagnostic tools before deploying synthetic datasets in safety-critical applications.
Measuring the Dependency Gap: Diagnosing Inter-Column Fidelity in Tabular Generative Models
arXiv:2607.21636v1 Announce Type: cross Abstract: Synthetic tabular data is valued for preserving not only each column's marginal distribution but the dependencies between columns -- structure that carries much of the discriminative signal for minority classes in imbalanced domains such as fraud and clinical risk. Yet the metrics most commonly used to certify synthetic tabular data are, we show, largely blind to inter-column dependency: a baseline that models every column independently (and therefore destroys all dependency) is judged indistinguishable from real data by the logistic-regression C2ST, and the pairwise Trend score is only partially sensitive. We introduce a dependency-aware fidelity diagnostic that decomposes a strong classifier two-sample test (XGB-C2ST) into marginal, dependency, and numerical-categorical cross components, anchored between a worst-case fully-factorized reference (all dependency destroyed) and a best-case real-data oracle. Applying it to a state-of-the-art flow-matching generator (TabbyFlow/EF-VFM), we find a real dependency gap that standard metrics miss; destroying dependency outright collapses minority-class utility, and the generator's residual gap carries a smaller, consistent utility cost. We then ask whether this gap reflects a structural limitation of mean-field generative objectives. It does not: consistent with recent recovery results for variational flow matching, the objective is asymptotically exact. Yet the gap is stubborn -- a 16x increase in model capacity does not close it -- pointing to the absence of direct dependency supervision rather than a capacity or structural limit. Consistent with this, and because the residual gap is higher-order, no cheap intervention closes it: not an in-model dependency mechanism, not post-hoc copula correction, and not the 16x capacity increase -- a caution for a field that assumes such fixes help.
Abstract:Synthetic tabular data is valued for preserving not only each column's marginal distribution but the dependencies between columns -- structure that carries much of the discriminative signal for minority classes in imbalanced domains such as fraud and clinical risk. Yet the metrics most commonly used to certify synthetic tabular data are, we show, largely blind to inter-column dependency: a baseline that models every column independently (and therefore destroys all dependency) is judged indistinguishable from real data by the logistic-regression C2ST, and the pairwise Trend score is only partially sensitive. We introduce a dependency-aware fidelity diagnostic that decomposes a strong classifier two-sample test (XGB-C2ST) into marginal, dependency, and numerical-categorical cross components, anchored between a worst-case fully-factorized reference (all dependency destroyed) and a best-case real-data oracle. Applying it to a state-of-the-art flow-matching generator (TabbyFlow/EF-VFM), we find a real dependency gap that standard metrics miss; destroying dependency outright collapses minority-class utility, and the generator's residual gap carries a smaller, consistent utility cost. We then ask whether this gap reflects a structural limitation of mean-field generative objectives. It does not: consistent with recent recovery results for variational flow matching, the objective is asymptotically exact. Yet the gap is stubborn -- a 16x increase in model capacity does not close it -- pointing to the absence of direct dependency supervision rather than a capacity or structural limit. Consistent with this, and because the residual gap is higher-order, no cheap intervention closes it: not an in-model dependency mechanism, not post-hoc copula correction, and not the 16x capacity increase -- a caution for a field that assumes such fixes help.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2607.21636 [cs.LG]
(or arXiv:2607.21636v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2607.21636 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 358 words
Related coverage
Constraint-Aware Synthetic Tabular Data Generation via Inter-Column Constraint Discovery with LLM Agents
arXiv · Aug 18, 2026
Beyond Distribution Matching: Semantics-Consistent Tabular Diffusion with Weak Semantic Priors
arXiv · Sep 16, 2026
DI Series: Serving Kimi-K3 MXFP4 Wide-EP Disaggregated on AMD Instinct MI300X and MI325X
AMD · Oct 8, 2026
Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement
Microsoft Research · Aug 11, 2026