KRR Group Meetings

Bigraphs originated as a model for mobile communicating agents. A bigraph models a set of entities which may: (1) have spatial relations (limited to one being inside another, e.g. a person in a room in a building) and also (2) communication links (e.g. one computer is linked via a network to many others, or two people are linked while talking on the phone).
In this talk I consider a simple model of spatial dynamics in which there are two kinds of region (think of land and water, or maybe infected and uninfected regions in some sense). Regions of a given type can merge and split as well as appear and disappear. A region of one type can be nested inside a region of the opposite type, but more complex spatial relations are not permitted.
I will show how bigraphs can be used to model the dynamic behaviour of such regions.

Qualitative approaches are a useful means to model spatial configurations and processes on an abstract level of detail. In this talk I will give an overview of my previous work in this area to encourage future collaborations and exchange of expertise.

I will give an introduction to work on (1) modeling qualitative spatial orientation calculi and (2) how such calculi can be applied for agent control in the domain of rule compliant vessel navigation. Due to abstraction of real world perceptions or merging of several knowledge bases inconsistent world models may arise. We tackled this problem by developing (3) a neighborhood-based approach for resolving such inconsistencies. Furthermore, I will touch how (4) spatial and terminological reasoning may support the architectural design process wrt. specific requirements.
Additionally, I will introduce work of my colleague L. Frommberger who (5) investigated how abstraction can be applied to reinforcement learning processes.

The idea of a rough set is that it provides a less detailed, or coarser, view of a set. The coarseness or granularity arises from a notion that some elements of the set are indistinguishable from each other. There have been a number of suggestions for how this topic could be applied to ontologies.

In this talk I will give a survey of some work on ontologies at different levels of detail. Several of the papers I will consider present proposals for some form of rough description logic and I will discuss how these relate to each other and what the practical applications might be.

For an autonomous agent to interact rationally within its environment, it must have knowledge of that environment. When we look at creating synthetic agents, how can this knowledge be acquired purely from its available sensors?

This talk gives an overview of one approach to this question, an approach that takes inspiration from the psychological theory of classical conditioning.

The success of knowledge extraction depends on many factors, such as the domain, the representativity of the corpus and the availability quallity of the linguistic resources, tools and algorithms.

In this talk I will discuss various aspects of nlp-based information extraction and knowledge modelling in different domains, with an emphasis on the legal domain. I will describe how certain types of information can be gathered from text, whereby both manual and automatic text analysis guides the modelling. The results can be e.g. transferred into a thesaurus, re-engineered into an ontology, or used for ontology enrichment and maybe further ones

Download link: presentation slides