Multiagent systems 1000-2D10SW1S

Autonomous agents and multi-agent systems or, more generally, multi-agent environments, represent a modern approach to design and implementation of complex and intelligent computational systems. An agent is a computational entity, such as a software program or a robot, capable of perceiving the environment and affecting it by performing actions, usually involving cooperation with human operators or other agents.

Actions of an agent are autonomous, and its intelligence is manifested in the capability for flexible and rational reactions to various, often unpredictable, situations that arise in the environment, as well as the ability to engaging in proactive behaviour, when necessary. Being a part of multi-agent system, an agent is an interactive entity that can engage in a teamwork with other agents and/or human users. Therefore the strength of multi-agent approach lies in complex patterns of interaction. The key
example is cooperation, allowing groups of agents to achieve common goals which are beyond reach of the individual members.

An assumption of autonomy of an agent requires introduction of mechanisms for resource allocation and decision making that stem from game theory, social choice theory, and mechanism design. Relying on the assumption of agents rationality, such mechanisms may lead to enforcing desired properties of the behaviour of agents or the effects of their interaction.

Topics of the seminar include, but are not restricted to the following:

1. Methods for reaching agreement in multi-agent systems:
- auctions,
- negotiations.

2. Knowledge representation and fusion
- rule-based query languages
- nonmonotonic reasoning

2. Group decision making
- voting and other procedures of preference aggregation
- database and knowledge base techniques.

3. Interaction patterns in multi-agent systems:
- group problem solving,
- methods of actions coordination.

4. Multi-agent communication languages

5. Ontologies, i.e. specification of the hierarchy of concepts and roles.

6. Modern programming environments and platforms for agent development (e.g. FIPA, JASON, JADE).

M.Sc. theses prepared in association with the seminar may be:

- practical (e.g. extending an existing system with a new functionality, extending or improving an existing negotiation mechanism), or
- theoretical (e.g. designing a formal model of a specific agent type, analysing or creating a resource allocation, preference aggregation mechanism, database querying algorithms dedicated for a particular domain),

During the seminars we will also review the most recent multi-agent literature.