Agent-Based IS

Motivation
Multi-agent systems (MAS) can typically become very complex and their behaviors can be hard to specify. Since, by definition, a MAS consists of a group of autonomous agents, one of the key challenges in designing a MAS is coordinating the actions of the agents. In a dynamic environment where the actions of the agents are also interdependent on each other, it is especially critical that the agents learn to adapt their actions to the actions of other agents. Furthermore, when designing agent systems it is impossible to foresee all the potential situations an agent may encounter and define behavioral repertoires and activities optimally in advance. Agents therefore have to learn from, and adapt to, their environment, especially in a multi-agent setting.
Until recently, research in the field of machine learning (ML) mainly concentrated on learning techniques and methods in single-agent or isolated-system settings. More and more, ML is being explored as a vital component to address challenges in multi-agent systems. For example, many application domains are envisioned in which teams of software agents or robots learn to cooperate amongst each other and with human beings to achieve global objectives. Learning may also be essential in many non-cooperative domains such as economics and finance, where classical game-theoretic solutions are either unfeasible or inappropriate.
At the same time, multi-agent learning poses significant theoretical challenges, particularly in understanding how agents can learn and adapt in the presence of other agents that are simultaneously learning and adapting. This is a fertile area of research that seems ripe for progress: the numerous and significant theoretical developments of the 1990s, in fields such as Bayesian, game-theoretic, decision-theoretic, and evolutionary learning, can now be extended to more challenging multi-agent scenarios.

Perspectives
We see two main streams of MAS research that incorporate learning: (1) MAS that incorporate learning into the agents so that the agents learn to adapt to their environment, and (2) MAS systems designed for learning tasks or for discovering new knowledge (for e.g., data mining).




	RA2: Agent Learning Overview Motivation Multi-agent systems (MAS) can typically become very complex and their behaviors can be hard to specify. Since, by definition, a MAS consists of a group of autonomous agents, one of the key challenges in designing a MAS is coordinating the actions of the agents. In a dynamic environment where the actions of the agents are also interdependent on each other, it is especially critical that the agents learn to adapt their actions to the actions of other agents. Furthermore, when designing agent systems it is impossible to foresee all the potential situations an agent may encounter and define behavioral repertoires and activities optimally in advance. Agents therefore have to learn from, and adapt to, their environment, especially in a multi-agent setting. Until recently, research in the field of machine learning (ML) mainly concentrated on learning techniques and methods in single-agent or isolated-system settings. More and more, ML is being explored as a vital component to address challenges in multi-agent systems. For example, many application domains are envisioned in which teams of software agents or robots learn to cooperate amongst each other and with human beings to achieve global objectives. Learning may also be essential in many non-cooperative domains such as economics and finance, where classical game-theoretic solutions are either unfeasible or inappropriate. At the same time, multi-agent learning poses significant theoretical challenges, particularly in understanding how agents can learn and adapt in the presence of other agents that are simultaneously learning and adapting. This is a fertile area of research that seems ripe for progress: the numerous and significant theoretical developments of the 1990s, in fields such as Bayesian, game-theoretic, decision-theoretic, and evolutionary learning, can now be extended to more challenging multi-agent scenarios. Perspectives We see two main streams of MAS research that incorporate learning: (1) MAS that incorporate learning into the agents so that the agents learn to adapt to their environment, and (2) MAS systems designed for learning tasks or for discovering new knowledge (for e.g., data mining). References To be added. Under construction. (Created by: rs, 10/24/03; last updated by: rs, 10/24/03.)