SAINT: Improved Neural Networks for Tabular Data via Row Attention and Contrastive Pre-Training

TL;DR

SAINT is a novel neural network architecture that employs row and column attention with contrastive pre-training, significantly enhancing performance on tabular data tasks and surpassing traditional gradient boosting methods.

Contribution

Introduces SAINT, a hybrid deep learning model with attention mechanisms and contrastive pre-training for improved tabular data analysis.

Findings

SAINT outperforms previous deep learning models on benchmark datasets.

SAINT surpasses gradient boosting methods like XGBoost, CatBoost, and LightGBM.

Contrastive pre-training benefits when labels are limited.

Abstract

Tabular data underpins numerous high-impact applications of machine learning from fraud detection to genomics and healthcare. Classical approaches to solving tabular problems, such as gradient boosting and random forests, are widely used by practitioners. However, recent deep learning methods have achieved a degree of performance competitive with popular techniques. We devise a hybrid deep learning approach to solving tabular data problems. Our method, SAINT, performs attention over both rows and columns, and it includes an enhanced embedding method. We also study a new contrastive self-supervised pre-training method for use when labels are scarce. SAINT consistently improves performance over previous deep learning methods, and it even outperforms gradient boosting methods, including XGBoost, CatBoost, and LightGBM, on average over a variety of benchmark tasks.