A Relational Model for One-Shot Classification of Images and Pen Strokes

We show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without relying on extensive data augmentation. Our study shows excellent results be achieved in which the is applied images, while building an efficient one-shot classifier top of raw strokes more challenging. The proposed classification performs matching pair inputs form local and pairwise attention. approach solves almost perfect accuracy image Omniglot challenge when combined Hungarian algorithm attains competitive same task characters represented as rotation-augmented strokes.