Development of Multimodal Pretraining and Representation Methods for Document Processing - Fujitsu Laboratories LTD