Cross-task Generative Augmentation of Multimodal Speech Features for Task-Incomplete MCI Detection | AMiner