This work tries to reduce the need for annotated data to design and train reasonably high performing neural networks for a given problem by using so-called few-shot learning, in combination with other techniques to further ...