In this paper, we propose
the LayoutLM to jointly model interactions between text and layout information across scanned document images, which is
beneficial for a great number of real-world document image understanding tasks such as information extraction from
scanned documents.