Geometric Mean Improves Loss For Few-Shot Learning

Few-shot learning (FSL) is a challenging task in machine learning, demandinga model to render discriminative classification by using only a few labeledsamples. In the literature of FSL, deep models are trained in a manner ofmetric learning to provide metric in a feature space which is wellgeneralizable to classify samples of novel classes; in the space, even a fewamount of labeled training examples can construct an effective classifier. Inthis paper, we propose a novel FSL loss based on \emph{geometric mean} to embeddiscriminative metric into deep features. In contrast to the other losses suchas utilizing arithmetic mean in softmax-based formulation, the proposed methodleverages geometric mean to aggregate pair-wise relationships among samples forenhancing discriminative metric across class categories. The proposed loss isnot only formulated in a simple form but also is thoroughly analyzed intheoretical ways to reveal its favorable characteristics which are favorablefor learning feature metric in FSL. In the experiments on few-shot imageclassification tasks, the method produces competitive performance in comparisonto the other losses.