Current students

In recent years, Knowledge Graphs have become ubiquitous, powering recommendation systems, natural language processing, and query answering, among others. Although KGs effectively represent structured data, manipulating such graphs is problematic at a practical level. For this reason, Representation Learning (RL) through Knowledge Graph Embedding (KGE) has gained increasing attention due to its unprecedented effectiveness in representing real-world structured information while preserving relevant properties. However, different KGE models leverage different aspects of the same KG, and, consequently, they benefit from different representations of the underlying data. Moreover, the increased complexity of data and models' structures makes the application development process longer and more error-prone than conventional machine learning, significantly increasing the probability of adopting sub-optimal solutions.
The end goal of this research project is to make Representation Learning on Knowledge Graphs faster and readily available to researchers and industry. To this aim, the project focuses on developing a framework to guide the implementation of efficient and effective KG-powered applications. The framework will support automatic subsetting of entities, relations, and features, as well as data-driven and task-driven model selection, in order to improve, at the same time, both the accuracy and the time requirements of different learning tasks on KGs. The framework will also transparently leverage heterogeneous computer architectures to furtherly boost the performance of Knowledge Graph Embedding models, without affecting the application development flow.