The goal of few-shot learning is to develop machine learning models which can learn to solve a task, typically a classification task, from very few labeled examples. For the field of relation extraction, in which the goal is to detect relations between entities (such as a person or organization) mentioned in a text, few-shot learning is particularly important: A core issue that makes relation extraction so challenging to evaluate, let alone tackle using machine learning models, is the fact that most relations only rarely occur at all. This means that in real world settings, we have few examples to train and evaluate on.
While especially for image classification tasks, a variety of different approaches have been developed and examined , few-shot relation extraction approaches  mostly make use of prototypical neural networks . The goal of the proposed thesis is to apply few-shot learning approaches, which have not yet been examined for relation extraction, to the few-shot relation extraction setting.
Hands-on experience in machine learning, no fear to implement neural network models (under guidance of the supervisors).