An Overview of Cooperative and Competitive Multiagent Learning

Hoen, Pieter Jan't; Tuyls, Karl; Panait, Liviu; Luke, Sean; Poutré, J.A.

doi:10.1007/11691839_1

Cited by 53 publications

(34 citation statements)

References 98 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Many real-life problems may be modeled with the help of the so-called complex dynamical systems (see, e.g., [2,10]) or, putting it in an other way, autonomous multiagent systems (see, e.g., [17]) or swarm systems (see, e.g., [29]). These are sets consisting of complex objects which are characterized by the persistent changes of parameters of their components over time, numerous relationships among the objects, the possibility of cooperation/competition among the objects and the ability of objects to perform more or less complicated actions (see, e.g., [3] for more details).…”

Section: Methods Of Behavioral Pattern Identificationmentioning

confidence: 99%

A Domain Knowledge as a Tool For Improving Classifiers

Bazan

Buregwa-Czuma

Jankowski

2013

Fundamenta Informaticae

View full text Add to dashboard Cite

This paper investigates the approaches to an improvement of classifiers quality through the application of a domain knowledge. The expertise may be utilizable on several levels of decision algorithms such as: feature extraction, feature selection, a definition of temporal patterns used in an approximation of the concepts, especially of the complex spatio-temporal ones, an assignment of an object to the concept and a measurement of the objects similarity. The domain knowledge incorporation results then in the reduction of the size of searched spaces. The work constitutes an overview of classifier building methods efficiently utilizing the expertise, worked out latterly by Professor Andrzej Skowron research group. The methods using domain knowledge intended to enhance the quality of classic classifiers, to identify the behavioral patterns and for automatic planning are discussed. Finally it answers a question whether the methods satisfy the hopes vested in them and indicates the directions for future development.

show abstract

Section: Methods Of Behavioral Pattern Identificationmentioning

confidence: 99%

A Domain Knowledge as a Tool For Improving Classifiers

Bazan

Buregwa-Czuma

Jankowski

2013

Fundamenta Informaticae

View full text Add to dashboard Cite

show abstract

“…18 In this respect, individual agents are computational procedures that perceive their environment, make inferences on the basis of the received percepts and their learned experience, and acts on their environment to reach predefined design goals. A full list properties associated with intelligent agents can be found in the work of Hoen et al 15 An agent in a MAS can be considered as an entity with an architecture comprising 2 fundamental components, namely the agent's hardware and the agent's software. In intelligent MASs, individual agents are required to be autonomous, which means learning capability through interactions with the environment as well as adapting to changes in the environment caused by agents' actions internally and the environments' dynamics externally.…”

Section: Multiagent Systemsmentioning

confidence: 99%

“…15 Basically, each agent gets a set of percepts from their environment, processes the percepts under light of their accumulated knowledge, and acts on the environment through their available operators to achieve a predefined design goal. 15 Basically, each agent gets a set of percepts from their environment, processes the percepts under light of their accumulated knowledge, and acts on the environment through their available operators to achieve a predefined design goal.…”

Section: Introductionmentioning

confidence: 99%

A multimetric and multideme multiagent system for multiobjective optimization

Tamouk

Acan

2018

Computational Intelligence

View full text Add to dashboard Cite

This article proposes a multiagent system consisting of a number of multiobjective metaheuristic agents (namely, multiobjective genetic algorithm, strength Pareto evolutionary algorithm, differential evolution, simulated annealing, and particle swarm optimization) working toward to extract optimal or very close-to-optimal Pareto fronts using multiple performance metrics in a sessionwise manner. At the beginning of each session, the main population is divided into a number of subpopulations, and each of them is assigned to a particular agent. The system runs in consecutive sessions such that, at the beginning of a session, agents start running after being assigned with a subpopulation and return the optimized subpopulations together with the corresponding set of nondominated solutions at the end of the session. There are 3 multiobjective assessment metrics in use, and a different metric is considered for each session to measure the success of each metaheuristic agent. The evaluation of individual agents using a particular assessment metric is used in 2 ways: first, the number of fitness evaluations for each agent is adjusted based on their performance; second, the subpopulation improved by an individual agent might be rejected on the basis of its evaluation score. At the end of each session, individual subpopulations are merged to get the updated main population, whereas individual sets of nondominated solutions are combined to form the global Pareto front. In addition to the individual multiobjective metaheuristic agents, the system also contains a number of coordination and synchronization agents that run the whole system toward its objectives. 1122

show abstract

“…Of course, it is known that Q-learning does not, in the general case, converge to equilibrium in 2-player repeated games [4,23,10]. However, there are a number of features that hold for the EI game matrix in the domains we study, which makes the specific situation special.…”

Section: Learning Payoffs In Lct Matrix Gamesmentioning

confidence: 99%

Adaptive multi-robot coordination: A game-theoretic perspective

Kaminka

Erusalimchik

Kraus

2010

2010 IEEE International Conference on Robotics and Automation

View full text Add to dashboard Cite

Multi-robot systems researchers have been investigating adaptive coordination methods for improving spatial coordination in teams. Such methods adapt the coordination method to the dynamic changes in density of the robots. Unfortunately, while their empirical success is evident, none of these methods has been understood in the context of existing formal work on multi-robot learning. This paper presents a reinforcement-learning approach to coordination algorithm selection, which is not only shown to work well in experiments, but is also analytically grounded. We present a reward function (Effectiveness Index, EI), that reduces time and resources spent coordinating, and maximizes the time between conflicts that require coordination. It does this by measuring the resource-spending velocity. We empirically show its success in several domains, including robots in virtual worlds, simulated robots, and physical AIBO robots executing foraging. In addition, we analytically explore the reasons that EI works well. We show that under some assumptions, spatial coordination opportunities can be modeled as matrix games in which the payoffs are directly a function of EI estimates. The use of reinforcement learning leads to robots maximizing their EI rewards in equilibrium. This work is a step towards bridging the gap between the theoretical study of interactions, and their use in multi-robot coordination.

show abstract

An Overview of Cooperative and Competitive Multiagent Learning

Cited by 53 publications

References 98 publications

A Domain Knowledge as a Tool For Improving Classifiers

A Domain Knowledge as a Tool For Improving Classifiers

A multimetric and multideme multiagent system for multiobjective optimization

Adaptive multi-robot coordination: A game-theoretic perspective

Contact Info

Product

Resources

About