OnEmbeddingsforNumericalFeatures inTabularDeepLearning

Unlike traditional models, e.g., MLP,these architectures mapscalar valuesofnumerical features tohigh-dimensional embeddings before mixing them inthemain backbone.