Few-shot learning is a learning method, in which in contrast to regular deep learning methods the amount of available training data is small. The assumption is that reliable algorithms can be built to achieve competitive performance to the models trained with abundant data.