HACNet: End-to-end learning of interpretable table-to-image converter and convolutional neural network

Takuya Matsuda, Kento Uchida, Shota Saito, Shinichi Shirakawa

Abstract: Highlights•We propose HACNet, a hard attention-based converter combined with CNN.•Different from existing methods, HACNet enjoys end-to-end learning of the converter.•The hard attention extracts the important features, which improves explainability.•The usage of template image helps us to make interpretable image from tabular data.•HACNet achieves higher accuracies in several tabular datasets over existing methods.