Research map / Label-efficient and robust learning / Self-supervised and contrastive
TabNet: Attentive Interpretable Tabular Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use sequential instance-wise sparse feature selection and reasoning | TabNet · Syn2 and Syn5 | Test AUC: TabNet .892 $\pm$ .004 vs INVASE .877 $\pm$ .003 (Syn2); .789 $\pm$ .009 vs .784 $\pm$ .005 (Syn5) | Helped |
| use sequential instance-wise sparse feature selection and reasoning | TabNet · Syn1, Syn3, Syn4, Syn6 | Test AUC: Syn1 .682 $\pm$ .005 vs .690 $\pm$ .006; Syn3 .897 $\pm$ .003 vs .902 $\pm$ .003; Syn4 .776 $\pm$ .017 vs .787 $\pm$ .00 | Mixed |
| use TabNet rather than AutoML Tables | Forest Cover Type · Forest Cover Type | Test accuracy (%): 94.95 -> 96.99 | Helped |
| use TabNet rather than rule-based classification | Poker Hand · Poker Hand | Test accuracy (%): 100.0 -> 99.2 | Hurt |
| increase TabNet model size from S to M or L | TabNet · Sarcos | Test MSE: 1.25 -> 0.28 -> 0.14 | Helped |
| use TabNet-M rather than sparse evolutionary MLP | Higgs Boson · Higgs Boson | Test accuracy (%): 78.47 -> 78.84 | Helped |
| use TabNet rather than CatBoost | Rossmann Store Sales · Rossmann Store Sales | Test MSE: 489.75 -> 485.12 | Helped |
| add masked-feature self-supervised pre-training | TabNet-M · Higgs, 1k training samples | Test accuracy (%): 57.47 $\pm$ 1.78 -> 61.37 $\pm$ 0.88 | Helped |
| add masked-feature self-supervised pre-training | TabNet-M · Higgs, 10k training samples | Test accuracy (%): 66.66 $\pm$ 0.88 -> 68.06 $\pm$ 0.39 | Helped |
| add masked-feature self-supervised pre-training | TabNet-M · Higgs, 100k training samples | Test accuracy (%): 72.92 $\pm$ 0.21 -> 73.19 $\pm$ 0.15 | Helped |
| reduce units to N_d=N_a=32 (ablation) | TabNet · Forest Cover Type | Test accuracy (%): 96.99 -> 94.99 (-2.00) | Hurt |
| reduce decision steps to 3 (ablation) | TabNet · Forest Cover Type | Test accuracy (%): 96.99 -> 96.22 (-0.77) | Hurt |
Closest papers by meaning
- InterpreTabNet: Distilling Predictive Signals from Tabular Data by Salient Feature Interpretation (ICML 2024 · similarity 0.90)
- Interpretable Mesomorphic Networks for Tabular Data (NeurIPS 2024 · similarity 0.90)
- TabM: Advancing tabular deep learning with parameter-efficient ensembling (ICLR 2025 · similarity 0.90)
- TabR: Tabular Deep Learning Meets Nearest Neighbors (ICLR 2024 · similarity 0.89)
- TabSTAR: A Tabular Foundation Model for Tabular Data with Text Fields (NeurIPS 2025 · similarity 0.89)
- Hybrid Autoencoders for Tabular Data: Leveraging Model-Based Augmentation in Low-Label Settings (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.