Background knowledge: Techniques are needed that 1) can handle large numbers of background predicates (Relevance), 2) can update theories structured in many levels (Revision), 3) can carry out predicate invention in deep-structured theories (Invention).
Complex Hypotheses: Techniques are needed that 4) can learn deep structured theories and optimise the choice of a set of clauses for a single predicate (Multi-Clause), 5) can handle long chains of relevant literals, connected by shared variables (Deep), 6) can better handle recursive hypotheses (Recursion), 7) can search efficiently in the presence of structural concepts expressed in complex clauses (Structure).
Built-in semantics: Techniques are needed that 8) better handle numbers (Numbers), 9) can express probabilistic constraints and definitions (Probabilities), 10) can learn and use constraints (Constraints), 11) work more efficiently through the use of built-in predicates and algorithms (Built-in).
Sampling issues. Techniques are needed that 12) can learn from large data sets (Large Data), 13) can learn from small data sets (Small Data), and 14) offer some reliability guarantees (Reliability). \end{itemize} The main methodology applied will be 1) to study the scientific problem starting from given application domains and data (provided by the end-user club), 2) to generalize away from the application, 3) to develop theory, techniques and implementations to cope with a specific problem, 4) to evaluate the developed framework on the application domains and data, and 5) to use the obtained feedback to re-iterate if necessary.
There is also a description of There is also a descriptions of international partners associated with the project.
Home Page