Multi-Agent Systems: a way of thinking
intelligence collectively
Abstract: Multi-agent systems (MAS) are a wide field
of research at the junction of several other fields in computer
science (such as artificial intelligence, software engineering
and distributed systems) and also other fields outside of computer
science like biology, social science, or business science.
MAS is today, indisputably, a suitable paradigm to build intelligent
systems where: 1) the basic entities, modelled as agents, have
local information or capabilities for solving the problems and,
thus, have only a partial view; 2) both the control and the knowledge
of the system are decentralised (intrinsically or for efficiency
and robustness reasons); and 3) the computation is asynchronous
and potentially distributed among a network of computers (e.g.
WEB applications, telecommunication software, diagnostic programs
or medical applications, etc. ).
The agents are considered to be autonomous entities (e.g. ants,
robots, drones, etc.). Their interactions lie within either cooperative
or competitive behaviours. That is, the agents can share a common
goal (cooperative agents), or they can pursue their own interests
(self-interested or competitive agents).
This talk will focus on MAS as a cognitive approach that helps
developing distributed, cooperative and intelligent systems where
intelligence is built collectively among agents. It is structured
around three main aspects of MAS:
- Cognition refers to intelligent skills of agents (autonomy,
learning, planning, reasoning, etc.) and endows agents with
reflexive behaviour, i.e. agents can act, observe and control
autonomously their actions.
- Interaction allows agents, viewed as intelligent and autonomous
entities, to exchange their knowledge (or beliefs), to solve
potential conflicts or to enhance synergy between their activities.
Interaction is the basic mechanism that supports more sophisticated
ones such as coordination, negotiation or consensus reaching.
- Concurrency enables the agents to achieve distributed tasks
both in cooperative and competitive cases. Concurrency provides
the essential and well established mechanisms (e.g. synchronous
or asynchronous communications, distributed observation, mobility,
etc.) that support the effective running of distributed MAS.
This talk will also outline the main domains of applications
where MAS technology brings an added-value while building distributed
and cooperative intelligent systems. The whole concepts and mechanisms
presented in this talk will be illustrated through two industrial
applications we developed in the domain of Aircraft simulation
and Telecommunication.
[ ]
|