W. Liu, D. Dubois, D. Bell, L. Godo, C. Sierra, J. Hong
In large scale, distributed data-driven and knowledge-driven intelligent systems, modelling and managing heterogeneous information from multiple sources and making use of expert/domain knowledge to assist decision making faces many challenges. These challenges include:
- Modelling: how to determine and select adequate theories/ formalisms to represent a variety of data, information, and knowledge; reasoning with such data/information and knowledge;
- Uncertainty/inconsistency: how to handle uncertainty, reliability, incompleteness, and inconsistency in data/information and knowledge;
- Fusion: how to merge or combine information and knowledge provided by multiple sources; the aim of such a fusion is to lay bare the reliable part of the available data, taking into account possible inconsistencies, while not introducing artificial precision;
- Change: how to reflect current available information in a dynamic environment by revising or updating agents' beliefs and knowledge: an intelligent system should account for the flux of new evidence, and learn from it;
- System development: how to design and evaluate robust, scalable, and fault tolerant computerized systems based on theoretical development in the above areas to meet the requirements of real-world applications.
Research in the Knowledge and Information Fusion group in KDE has been primarily carried out to meet these challenges.
Our research starts with establishing sound theoretical foundations for modelling and managing heterogeneous information and knowledge.
- examining and comparing different information/knowledge representation theories, especially under uncertainty and inconsistency;
- discovering general principles governing the fusion of information/ knowledge, regardless of what theory is chosen to model such information or knowledge;
- developing fusion or merging algorithms and strategies for combining information and knowledge from different sources;
- developing revision or updating strategies in dynamic systems to take into account new information or the changes of environment, or newly discovered knowledge.
We actively apply our theoretical findings to real-world applications. Application areas include belief fusion and revision in multi-agent systems, sensor information fusion and event reasoning in static and mobile environments for decision making in large sensor networks (especially for physical/cyber security, infrastructure security, and connected health), knowledge-driven risk analysis and requirement decisions in manufacturing processes, and semi-structured data fusion.
The deployment of open distributed systems is increasing rapidly. The advances of network technologies are spawning a surge of application domains: ambient intelligence, cloud computing, service oriented computing, sensor networks, or virtual organisations, just to mention a few. These applications are composed of a wealth of physical devices, software components and, frequently, humans. The overall co-ordination of these elements presents a tremendous challenge because of the lack of centralised control, the openness of the system, and the complexity and volatility of the environment.
We investigate the use of multi-agent technologies to model such systems. A multi-agent system is a community of autonomous agents that is situated in an environment. The particular technologies we are mostly interested in are:
- Electronic Institutions. Communities typically choose to regulate the behaviour of their constituent human, or artificial, agents; for example, a regulation to drive vehicles on the left side of the road. A multi-agent system together with a regulatory mechanism for monitoring, and possibly restricting or policing, agent behaviour that breaches the adopted regulations is known as an electronic institution. In this context, we are interested in the theory of normative systems and in developing support tools for electronic institutions to explore their usage in different application domains.
- Agent architectures. Agents participating in electronic institutions are decision makers. The complex decisions they have to face include how to co-ordinate with other agents in order to reach agreements for their mutual benefit. In particular we investigate the use of graded BDI architectures to underpin negotiation and planning algorithms.
- Modelling and reasoning about actions in noisy environments. In many situations, agents have to coordinate, plan and take decisions in noisy environments where only incomplete and uncertain information may be at hand. We aim at investigating (both qualitative and quantitative) models for decision making about actions for planning and negotiation in very poor information scenarios, but still able to produce meaningful outputs. In particular, foundations for generalized models for uncertainty (e.g. probabilistic information, fuzzy information) are being developed, as well as corresponding decision theories to be investigated.