ProtoNAM: Prototypical Neural Additive Models for Interpretable Deep Tabular Learning