Using KullBack-Liebler Divergence Based Meta-learning Algorithm for Few-Shot Skin Cancer Image Classification: Literature Review and a Conceptual Framework

Meta-learning often termed "learning to learn," seeks to construct models that can swiftly learn new information or adapt to new situations using only a few training data samples. Meta-learning is distinct from traditional supervised learning. The model is required to recognise training data before generalising to unknown test data in traditional supervised learning. Meta-learning, on the other hand, has only one goal: to learn. Few-shot makes predictions based on a limited amount of data using a prototypical network that it learns using a metric space in which classification is done by computing distances between prototype representations of each class. The Kullback-Leibler Divergence (KLD) a non-symmetric distance measure between two probability distributions, No and q (x) is used in the meta-learning process of few-shot leaning to compute the distance between the prototype class to query cancerous image. This study seeks to investigate the usefulness of KLD in cancer image classification to assist medical practitioner in early diagnosis of the dreadful disease.