Sub-One Shot Learning is a learning paradigm that addresses extremely low-label scenarios, where the number of labeled data points is far smaller than the number of classes. In such settings, some classes have no labeled samples at all, making it difficult to generalize using conventional supervised learning approaches. Therefore, the model must learn shared representations and structural relationships among classes from the limited labeled information available.

• Problem Insufficient supervision when training samples are far fewer than target classes.
• Approach Learning shared representations and relationships among classes.
• Method Information augmentation leveraging label synergy and transfer learning.
• Extension Aligning distributions via domain adaptation.
• Goal Ensuring stable generalization performance under extremely low-label conditions