促进跨模态表示以提升生理信号的多模态基础模型
原文英文，约400词，阅读约需2分钟。发表于：。Many healthcare applications are inherently multimodal, involving several physiological signals. As sensors for these signals become more common, improving machine learning methods for multimodal...
本文探讨了在PhysioNet 2018数据集上预训练多模态模型的挑战，强调跨模态重建目标对多模态训练的重要性。研究发现，模态丢弃能提升下游任务性能，最终模型的表示更加跨模态和时间对齐，验证了多模态基础模型在健康数据中的有效性。
