Logic, Data Examples, and Learning

About

In logic, data examples are useful when a logical formula must be synthesized or communicated. This includes tasks such as the reverse engineering of logical queries, debugging and refinement of formal specifications, as well as various forms of learning.
This course provides a uniform introduction to the use of data examples in logic, covering logical formalisms that range from propositional logic to first-order logic (including conjunctive queries and description logics), and addressing the following questions:
* When, and to what extent, can a logical formula be described by a small number of data examples. ("From Formulas to Data Examples")
* How to construct a fitting formula from a set of data examples? When more than one fitting formula exists, which one should be preferred? ("From Data Examples to Formulas")
* When is a fitting formula likely to generalize from input examples to unseen data examples, and how many input examples are required to achieve this? ("PAC Learning")
* In interactive design and refinement systems, data examples provide a means of communication between the system and the user. This can be modeled as an interactive learning setting with a learner and an oracle. When is efficient interactive learning of this type possible?

Course leader

Balder Ten Cate and Carsten Lutz

Target group

Students

Fee info