Transformer-based neural networks for tabular data are highly susceptible to backdoor attacks, even with minimal feature value alterations.
