The problem of theory formation from observed phenomena has been the focus of many researchers in many disciplines including biology, medicine, economics, criminology and computer science. The acknowledged complexity of building explanatory theories increases significantly when the available data is incomplete, inconsistent or subject to unpredicted changes. In these cases, the knowledge extraction process has to be robust to noise, able to integrate domain-specific heuristics for filtering out errors in observations, and to control additional bias induced by the incompleteness of the observations.
Inductive logic programming is a subfield of AI that is concerned with generating hypotheses from given observations (i.e., symbolic learning), as well as refining and revising existing theories. It has been applied to a wide range of inference problems such as business process modelling and software requirements elaboration. However, a significant impediment for its use in large-scale applications is the computational demand imposed by searching for hypothesis in large solution spaces.
This studentship will focus on developing distributed inductive logic programming theories, algorithms and tools to enable symbolic learning within the context of large-scale problems. The project will offer an opportunity to conduct inter-disciplinary work with researchers in domains such as software engineering, crime science and bioinformatics.