Transformer is all you need for multimodal multitask learning using a unified transformer.

该论文提出了一种名为UniT的方法：一种统一的Transformer编码器-解码器体系结构。这种方法能在单一模型架构中整合了多种任务与领域的学习过程，并能取得该领域内的最佳性能。代码开源工作已启动！

Transformers have emerged as the pinnacle of machine learning innovation, offering a unified solution for diverse tasks across multiple domains. A novel framework for multimodal and multitask learning has been developed, leveraging a unified transformer architecture to achieve exceptional performance in various applications.

• 所属机构：Meta（Research lab）
• 官方主页：https://mmf.sh/
• 论文下载链接：https:/