Abstract: Few-shot learning offers a promising solution for computational pathology by alleviating the reliance on large an-notated datasets, but faces challenges from the high redundancy in whole ...