While this excels at human understanding of natural language tasks, it seems generalizable across problem domains. Biomedical microscopy images for disease dete

While this excels at human understanding of natural language tasks, it seems generalizable across problem domains. Biomedical microscopy images for disease detection, for example. The "cross-modal" representation works! Just as effectively as it does for single modes such as image, text and sound. A very interesting finding ;)