Few-shot Named Entity Recognition (NER) systems are designed to identify new categories of entities with a limited number of labeled examples. A major challenge encountered by these systems is overfitting. particularly pronounced in comparison to tasks with ample samples. This overfitting predominantly stems from spurious correlations. https://www.soiebiologique.com/