FT-Transformer and tuned residual baselines for tabular learning

What changed

Compared tabular deep architectures under controlled splits/tuning and introduced feature tokenization of numerical and categorical inputs followed by Transformer layers over one object's features.

Scientific scope

Compared tabular deep architectures under controlled splits/tuning and introduced feature tokenization of numerical and categorical inputs followed by Transformer layers over one object's features. The comparison deliberately omits several model-agnostic improvements and covers a stated dataset regime. Its conclusion is not universal neural superiority: boosted trees and deep networks lead on different datasets. The simple residual baseline and Transformer adaptation are bundled, not credited with inventing ResNets or attention.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The comparison deliberately omits several model-agnostic improvements and covers a stated dataset regime. Its conclusion is not universal neural superiority: boosted trees and deep networks lead on different datasets. The simple residual baseline and Transformer adaptation are bundled, not credited with inventing ResNets or attention.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Revisiting Deep Learning Models for Tabular Data

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction