IDEAL: Independent Domain Embedding Augmentation Learning