Improving Tabular Data Contrastive Learning via Class-Conditioned and Feature-Correlation Based Augmentation

Incorporating class information and feature correlation structures into the tabular data augmentation process can improve the performance of contrastive learning for downstream classification tasks.