Learning Enhanced Representations for Tabular Data via Neighborhood Propagation

Recently, as shown by Papernot and McDaniel (2018), feeding in neighbors of the target as input can improve the robustness of data representations and help the model generalize better to out-of-distribution samples.