TabNet is a novel deep learning architecture designed for tabular data, aiming to achieve high performance and interpretability. It uses sequential attention to select the most salient features at each decision step, enabling efficient learning and interpretability. TabNet outperforms other tabular learning models on various datasets and provides local and global interpretability through feature importance masks. Additionally, TabNet demonstrates significant improvements in performance when using unsupervised pre-training for tabular data, leveraging the abundance of unlabeled data. The architecture is flexible, integrates well with end-to-end learning, and is applicable to a wide range of tabular datasets, including synthetic and real-world datasets.TabNet is a novel deep learning architecture designed for tabular data, aiming to achieve high performance and interpretability. It uses sequential attention to select the most salient features at each decision step, enabling efficient learning and interpretability. TabNet outperforms other tabular learning models on various datasets and provides local and global interpretability through feature importance masks. Additionally, TabNet demonstrates significant improvements in performance when using unsupervised pre-training for tabular data, leveraging the abundance of unlabeled data. The architecture is flexible, integrates well with end-to-end learning, and is applicable to a wide range of tabular datasets, including synthetic and real-world datasets.