Multiagent system

Lecture

Multi-agent system (MAS, English Multi-agent system ) is a system formed by several interacting intelligent agents. Multi-agent systems can be used to solve such problems that are difficult or impossible to solve with the help of a single agent or a monolithic system ( English ). Examples of such tasks are online trading ^[1] , emergency response ^[2] , and social structure modeling ^[3] .

Overview

In a multi-agent system, agents have several important characteristics ^[4] :

Autonomy : agents, at least partially, independent
Limited view : none of the agents have an idea about the entire system, or the system is too complex for knowledge of it to have practical application for the agent.
Decentralization : there are no agents controlling the entire system ^[5]

Typically, software agents are explored in multi-agent systems. However, robots, people, or teams of people can also be part of a multi-agent system. Also, multi-agent systems can contain mixed commands.

In multi-agent systems, self-organization and complex behavior can occur even if the behavior strategy of each agent is fairly simple. This is the basis of the so-called swarm intelligence.

Agents can share their knowledge using some special language and obeying the established rules of "communication" (protocols) in the system. Examples of such languages are Knowledge Query Manipulation Language (KQML) and FIPA's Agent Communication Language (ACL).

The study of multi-agent systems

The study of multi-agent systems is associated with solving artificial intelligence problems.

Topics for research within the MAS:

knowledge, desires and intentions (BDI),
cooperation and coordination,
organization,
communication,
matching,
distributed solution
distributed problem solving
multi-agent training
reliability and fault tolerance

Paradigms of multi-agent systems

Many MACs have computer implementations based on step-by-step simulation modeling. MAC components usually interact through a weight matrix of queries,

  Speed-VERY_IMPORTANT: min = 45 mph, 
  Path length-MEDIUM_IMPORTANCE: max = 60 expectedMax = 40, 
  Max-Weight-UNIMPORTANT 
  Contract Priority-REGULAR

and the response matrix,

  Speed-min: 50 but only if weather sunny,  
  Path length: 25 for sunny / 46 for rainy
  Contract Priority-REGULAR
  note - ambulance will wait

The “Request - Answer - Agreement” model is a common occurrence for MAS. The scheme is implemented in several steps:

First, everyone is asked a question like: "Who can help me?"
to which only the “capable” answer “I can, for such and such a price”
in the end, the “agreement” is established

The last step usually requires several more (smaller) acts of information exchange. This takes into account other components, including the “agreements” already reached and environmental constraints.

Another commonly used paradigm in MAC is “pheromone”, where components “leave” information for the next in line or nearest component. Such "pheromones" can evaporate with time, that is, their values may change with time.

Properties

MASs also belong to self-organizing systems, since they look for an optimal solution of the problem without external intervention. An optimal solution is a solution for which the least amount of energy is spent in conditions of limited resources.

The main advantage of the MAS is flexibility. Multi-agent system can be added and modified without rewriting a significant part of the program. Also, these systems are self-healing and resilient due to an adequate supply of components and self-organization.

MAC application

Multi-agent systems are used in our life in graphic applications, for example, in computer games. Agent systems have also been used in films ^[6] . The theory of MAS is used in composite defense systems. Also, MAS are used in transport, logistics, graphics, geographic information systems and many others. Multi-agent systems have proven themselves in the field of network and mobile technologies to ensure automatic and dynamic load balancing, extensibility and self-healing ability.