TabDeco pairs SAINT-style attention with SwitchTab-style feature decoupling and six contrastive losses, but its claim of consistently beating gradient boosting is contradicted by its own results.
Some recent advances in design of bayesian binomial reliability demonstration tests
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
TabDeco: A Comprehensive Contrastive Framework for Decoupled Representations in Tabular Data
TabDeco pairs SAINT-style attention with SwitchTab-style feature decoupling and six contrastive losses, but its claim of consistently beating gradient boosting is contradicted by its own results.