KRR Group Meetings

Qualitative spatial reasoning is a complex computational problem. Over the past 20 years there has been a lot of progress in enabling efficient reasoning for some of the well-known qualitative spatial calculi.
However, applications often require a customized spatial representation and existing theoretical results usually cannot be transferred to a new spatial representation.

In this talk I show how users of spatial information can easily develop their own qualitative spatial representation for a given application, and easily develop efficient spatial reasoning algorithms for their new representation.
Previously, this required a complicated technical analysis that could only be performed by a small number of experts. Using results and algorithms developed by the speaker, this task can now be done without any expertise in how to theoretically analyze spatial calculi.

Direction relations between extended spatial objects are important commonsense knowledge. Skiadopoulos proposed a formal model for representing direction relations between compound regions (the finite union of simple regions), known as SK-model. It perhaps is currently one of most cognitive plausible models for qualitative direction information, and has attracted interests from artificial intelligence and geographic information system. Due to the asymmetry and the massive number of basic directions, the inverse and composition operations in SK-model become extraordinary complex.

Here a novel method for the inverse and composition are proposed. Basing the concepts of smallest rectangular directions and its original directions, it transforms the inverse and composition of basic cardinal direction relations into the inverse and composition of rectangle relations ( planar extension of Allen's interval relations ).
Comparing with existing methods, this algorithm has quite good dimensional extendibility, that is, it can be easily transferred to the tri-dimensional space with a few modifications. Also it gives the hints to combine the temporal information with directions.

Humans tend to interpret a temporal series of geographical spatial data in terms of geographical processes. They also often ascribe certain properties to processes (e.g. a process may be said to accelerate). Given a spatial region of observation, distinct properties may be observed in different subregions and at different times, which causes difficulties for humans to identify them. The conceptualisation of geographical features and their correlation with geographical phenomena may provide a human like approach to analyse large spatio-temporal datasets.

This paper presents a representational model and a reasoning mechanism to analyse evolving geographical features and their relationship to geographical processes. The proposed approach comprises methods of relating occurrences of geographical events to geographical processes which is said to proceed over time. We introduce an initial set of properties which can be associated with several geographical processes. We consider this as a first step towards a more general model for representing and reasoning about geographical processes.

Paper going to be presented at the 4th International Conference on Geospatial Semantics (Geos 2011)

In order for an autonomous agent to interact rationally within its environment, it must have knowledge of that environment. Given that the wealth of knowledge that even small children evidently quickly acquire, it is infeasible for an agent to be directly encoded with much, if any, knowledge about the real world. This means that it would be best to instead imbue the agent with the ability to learn the knowledge for itself.

Given the non-triviality of the problem of programming an agent with this ability, this paper looks at a system that qualitatively replicates one of the main psychological processes that biological agents use to learn about their environment, that of classical conditioning. Initial testing of the system shows results that are inconclusive but are encouraging. This leads to the conclusion that further work is needed to ascertain the utility of the approach.

Undergraduate students will give out a brief presentation of the Final Year Projects they have been working on since January. These projects are all centred around Knowledge Representation and Reasoning topics: granularity in ontologies, improving automatic questionnaires, online games problem solving and similarity relations among colours.

The formal framework known as Defeasible Logic is used to represent argumentation. We show that, given a semantics based upon the notion of Individual Interpretation of a spatial concept, defeasible logic can be used to model vagueness in spatial reasoning.