TabNet: Attentive Interpretable Tabular Learning

TL;DR

TabNet is a novel deep learning architecture for tabular data that combines attention-based feature selection with interpretability, outperforming existing models and introducing self-supervised learning for tabular datasets.

Contribution

It introduces TabNet, a new interpretable neural network architecture with sequential attention, and demonstrates the effectiveness of self-supervised learning for tabular data.

Findings

TabNet outperforms other models on various tabular datasets.

It provides interpretable feature attributions and insights.

Self-supervised learning significantly improves performance with unlabeled data.

Abstract

We propose a novel high-performance and interpretable canonical deep tabular data learning architecture, TabNet. TabNet uses sequential attention to choose which features to reason from at each decision step, enabling interpretability and more efficient learning as the learning capacity is used for the most salient features. We demonstrate that TabNet outperforms other neural network and decision tree variants on a wide range of non-performance-saturated tabular datasets and yields interpretable feature attributions plus insights into the global model behavior. Finally, for the first time to our knowledge, we demonstrate self-supervised learning for tabular data, significantly improving performance with unsupervised representation learning when unlabeled data is abundant.