This repo contains a couple different experiments to quantify how an imbalanced or few-shot dataset will impact a model.
The purpose of this experiement is to quantify the impact of the imbalance. This isn't designed to be a SOTA model for solving the imbalance.
The dataset used for this experiment is a modified verion of MNIST where the minority class is limited to either 3 or 50 training examples.