Analysis of Transformer Model Applications.

Since the emergence of the Transformer, many variations of the original architecture have been created. Revisions and taxonomies have appeared that group these models from different points of view. However, no review studies the tasks faced according to the type of data used. In this paper, the modifications applied to Transformers to work with different input data (text, image, video, etc.) and to solve disparate problems are analysed. Building on the foundations of existing taxonomies, this work proposes a new one that relates input data types to applications. The study shows open challenges and can serve as a guideline for the development of Transformer networks for specific applications with different types of data by observing development trends.