Prototype Learning
Overview
Prototype learning studies how representative examples, feature centers, or structured sets of prototypes can organize data and guide predictions. Common approaches include metric-based few-shot networks, clustering and contrastive learning, memory banks, and hierarchical or probabilistic prototypes; prototypes also help adapt pretrained models across domains and modalities. The methods are used in settings such as image and point-cloud recognition, medical segmentation, anomaly detection, and federated or continual learning, often when labels are scarce or explanations matter. A central challenge is making prototypes diverse, stable, and faithful to model decisions while handling uncertainty, class imbalance, and shifts between training and test data.
Latest papers
88 papers10 papers in the last four weeks, up 67% on the four weeks before. 0.1% of all new papers.