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.
| Subjects: | Machine Learning (cs.LG); Machine Learning (stat.ML) |
| Cite as: | arXiv:1908.07442 [cs.LG] |
| (or arXiv:1908.07442v5 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.1908.07442 arXiv-issued DOI via DataCite |
Submission history
From: Sercan Arik [view email]
[v1]
Tue, 20 Aug 2019 15:46:53 UTC (1,030 KB)
[v2]
Wed, 28 Aug 2019 20:37:43 UTC (1,030 KB)
[v3]
Thu, 26 Sep 2019 00:59:17 UTC (1,031 KB)
[v4]
Fri, 14 Feb 2020 18:37:55 UTC (1,150 KB)
[v5]
Wed, 9 Dec 2020 05:00:33 UTC (1,149 KB)