CTTVAE: Latent Space Structuring for Conditional Tabular Data Generation on Imbalanced Datasets
Milosh Devic, Jordan Gierschendorf, David Garson · 2026
Read CTTVAE: Latent Space Structuring for Conditional Tabular Data Generation on Imbalanced Datasets by Milosh Devic, Jordan Gierschendorf, David Garson (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03641.txt.
Description
Synthetic data fails exactly where it matters most when the rare class is the important one, like fraud or disease. Milosh Devic and 2 co-authors introduce CTTVAE in a 2026 paper, a conditional transformer-based variational autoencoder for tabular data. The target setting is severe class imbalance, where most generators ignore the minority or produce useless samples.
Two mechanisms do the work. A class-aware triplet margin loss reshapes the latent space for tighter, better-separated classes, and a training-by-sampling strategy shows the model underrepresented groups more often. Across six real-world benchmarks, the combined system achieves the best downstream results on minority classes, sometimes beating models trained on the original data. Ablations credit both components for the gains.
Catalog record
| CREATOR | Milosh Devic, Jordan Gierschendorf, David Garson |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03641