Knowledge Graphs are collections of factual triplets, where each triplet (h, r, t) represents a relation r between a head entity h and a tail entity t.
Knowledge Graphs are widely used in a variety of applications such as question-answering, information retrieval, recommender systems, and natural language processing. Since knowledge graphs are usually incomplete, a fundamental problem is predicting the missing links.
However, a large portion of Knowledge Graph relations are long-tail. In other words, they have very few instances. But intuitively, the fewer training triples that one relation has, the more Knowledge Graph completion techniques could be of use. Therefore, it is crucial for models to be able to complete relations with limited numbers of triples.

Our work focuses on predicting missing relations in Knowledge Graphs with very few examples, what is known as Few-Shot Link Prediction in Knowledge Graphs. For that purpose we leverage Gradient-Based Meta-Learning methods which are efficient in fast adaptation of Neural Networks making them able to learn faster and better with very few samples.
In this talk, we will discuss the possibility of merging these widely used Meta-Learning methods with state-of-the-art Knowledge Graph embedding methods capable of capturing relation patterns such as symmetry, asymmetry, inversion, composition and transitivity. We believe being able to model these patterns is a bare minimum for having a good enough relation representation and a necessity for making good few-shot predictions.