Evolutionary design of agent-based simulation experiments

Citeseer

In this article we present EcoSFERES, a framework that helps applying learning and evolutionary algorithms to the design of multi-agent systems (MAS). This framework includes tightly linked abstractions for evolutionary computing and multi-agent based simulations, and additionally, agent learning policies can also be included and reused straightforwardly. In particular, EcoS-FERES makes it possible to implement a variety of learning and evolutionary techniques in order to compare their results in identical conditions, and it also facilitates important changes in the simulation setup. The main added values of this article are, on the one hand, to introduce the first development framework that provides a generic way of using evolutionary techniques for the design of MAS, and, on the other hand, to show that the tuning of MAS can be facilitated by a well designed development framework. EcoSFERES is an evolution of SFERES , which is the acronym of "SFERES is a Framework Enabling Research on Evolution and Simulation"

2019

In this paper, we introduce a new agent-based method to build a decision-aid tool aimed to improve policy design. In our approach, a policy is defined as a set of levers, modelling the set of actions, the means to impact a complex system. Our method is generic, as it could be applied to any domain, and be coupled with any agent-based simulator. We could deal not only with simple levers (a single variable whose value is modified) but also complex ones (multiple variable modifications, qualitative effects, ...), unlike most optimization methods. It is based on the evolutionary algorithm CMA-ES, coupled with a normalized and aggregated fitness function. The fitness is normalized using estimated Ideal (best policy) and Nadir (worst policy) values, these values being dynamically computed during the execution of CMA-ES through a Pareto Front estimated with the ABM simulation. Moreover, to deal with complex levers, we introduce the FSM-branching algorithm, where a Finite State Machine (FSM...

Journal of Computational Social Science

Agent-based modeling is a well-established discipline today with a rich and vibrant research community. The field of evolutionary computation (EC) is also well recognized within the larger family of computational sciences. In the past decades many agent-based modeling studies of social systems have used EC methods to tackle various research questions. Despite the relative frequency of such efforts, no systematic review of the use of evolutionary computation in agent-based modeling has been put forth. Here, we review a number of prominent agent-based models of social systems that employ evolutionary algorithms as a method. We comment on some theoretical considerations, the state of current practice, and suggest some best practices for future work.

Procedia Computer Science, 2014

This paper focuses on parameter search for multi-agent based models using evolutionary algorithms. Large numbers and variable dimensions of parameters require a search method which can efficiently handle a high dimensional search space. We are proposing the use of complexification as it emulates the natural way of evolution by starting with a small constrained search space and expanding it as the evolution progresses. We examined the effects of this method on an EA by evolving parameters for two multi-agent based models.

2008 SICE Annual Conference, 2008

Multi-agent systems have become more and more important in many aspects of computer science such as distributed artificial intelligence, distributed computing systems, robotics, artificial life, etc. Based on the belief that any complex system is more than the sum of its individual elements (Holand, 1999, Morgan, 1923, Morowitz, 2002), multi-agent systems introduce the issue of the emergence of behavior through interactions between agents (Forrest, 1991). Accordingly, a coordinated behavior needed to archive complex tasks might emerge in multi-agent systems from relatively simple defined interactions between agents. An agent is a virtual entity that can act, perceive the proximity of its environment and communicate with others; it is autonomous and has the ability to achieve its goals. Multiagent systems contain a world (environment), entities (agents), relations between the entities, a way the world is perceived by the entities, a set of operations that can be performed by the entities and changes in the world as a result of these actions. The main application areas of multi-agent systems are problem solving, simulation, collective robotics, software engineering, and construction of synthetic worlds (Ferber, 1999). Considering the latter application area and focusing on the autonomy of agents and the interactions that link them together (Parunak, Van, Brueckner, Fleischer & Odell, 2002), the following important issues can be raised: What is the minimum amount of perception information needed by agents in order to perceive the world? How can agents cooperate? What are the methods, and what are the lower bounds of communications, required for them to coordinate their actions? What architecture should they feature so that they can achieve their goals? What approaches can be applied to automatically construct their functionality, with the quality of such a design being competitive with human-handcrafted design? These issues are of special interest, since the aim is to create multi-agent systems, which are scalable, robust, flexible, and able to adapt to changes automatically. These features of multi-agent systems are believed to be particularly important in real world applications where the approaches to construct synthetic worlds can be viewed as practical methods, techniques towards creating complex "situation-aware" multi-computer, multivehicle, or multi-robot systems based on the concepts of agents, communication, cooperation and coordination of actions. The purpose of our research is an automatic design of the coordinated behavior among agents. Particularly, we are interested in the robustness of the coordinated behavior to some www.intechopen.com Multi-Agent Systems-Modeling, Interactions, Simulations and Case Studies

2012

This paper develops a multiobjective evolutionary algorithm on the cloud computing environment to help planners solve multiobjective problems more efficiently and effectively. The cloud environment is emulated as a virtualized biological world with several isolated regions. The main population initially continues evolutionary processes as the most widely-known evolutionary algorithms. To yield both exploration and exploitation, two processes, such as migration and interaction, are deployed. In the process of migration, local optimal solutions can migrate to form new populations so that the search space can be expanded. To overcome the disadvantages in isolated evolutionary algorithms, the individuals in different populations will interact stochastically in the interaction process. Taking the advantage of cloud computing environment, planners can take less effort on deploying both computation power and storage space. Instead, the planners can focus on the design of encoding, crossove...

Evolutionary Computation, 2009

Evolutionary algorithms are heuristic techniques for finding (sub)optimal solutions for hard global optimization problems. Evolutionary algorithms may be also applied to multimodal and multi-objective problems (for example compare ). In these cases however some special techniques must be used in order to obtain multiple high-quality solutions. Most important of these mechanisms are techniques that maintain population diversity because we are interested in finding the whole set of solutions-it would be a set of nondominated solutions in the case of multi-objective optimization (all notions and ideas of multiobjective optimization may be found in ) and a set of individuals located within basins of attraction of different local optima in the case of multi-modal optimization problems. Agent-based evolutionary algorithms result from mixing two paradigms: evolutionary algorithms and multi-agent systems. Two approaches are possible when we try to mix these two paradigms. In the first one we can use agent-based layer of the computing system as a "manager" of the computations. In this case each agent has sub-population of individuals inside of it. Agent tries to utilize computational resources in a best way-it observes the computational environment and tries to migrate to nodes which have free computational resources. In this approach evolving individuals are processed with the use of standard evolutionary algorithm. In the second approach individuals are agents which live within the environment composed of computational nodes, compete for resources, reproduce, die, observe the environment and other agents, communicate with other agents, and can change the environment. The selection is realized in the decentralized way: there are some resources defined in the system and "worse" agents (which have "worse" solutions encoded within their genotypes) give some amount of their resources to "better" agents. These resources are needed for all activities, like reproduction, migration, etc. When an agent runs out of resources it dies and is removed from the system. The example of the second approach is evolutionary multi-agent system (EMAS) model , and co-evolutionary multi-agent system (CoEMAS) model , which additionally allows for existence of many species and sexes of agents-it is also possible to define interactions between them and introduce coevolutionary interactions.

Lecture Notes in Computer Science, 2007

This paper addresses the problem regarding the parameter exploration of Multi-Agent Based Simulation for social systems. We focus on the principles of Inverse Simulation and Genetics-Based Validation. In conventional artificial society models, the simulation is executed straightforwardly: Initially, many micro-level parameters and initial conditions are set, then, the simulation steps are executed, and finally the macro-level results are observed. Unlike this, Inverse Simulation executes these steps in the reverse order: set a macro-level objective function, evolve the worlds to fit to the objectives, then observe the micro-level agent characteristics. Another unique point of our approach is that, using Genetic Algorithms with the functionalities of multi-modal and multi-objective function optimization, we are able to validate the sensitivity of the solutions. This means that, from the same initial conditions and the same objective function, we can evolve different results, which we often observe in real world phenomena. This is the principle of Genetics-Based Validation.

Computer Science, 2016

Computing applications such as metaheuristics-based optimization can greatly benefit from multi-core architectures available on modern supercomputers. In this paper, we describe an easy and efficient way to implement certain population-based algorithms (in the discussed case, multi-agent computing system) on such runtime environments. Our solution is based on an Erlang software library which implements dedicated parallel patterns. We provide technological details on our approach and discuss experimental results.

Evolutionary Computation, 2000

Holland's Adaptation in Natural and Artificial Systems largely dealt with how systems, comprised of many self-interested entities, can and should adapt as a whole. This seminal book led to the last 25 years of work in genetic algorithms (GAs), and related forms of evolutionary computation (EC). In recent years, the expansions of the Internet, other telecommunications technologies, and other large scale networks, have led to a world where large numbers of semi-autonomous software entities (i.e., agents) will be interacting in an open, universal system. This development cast the importance of Holland's legacy in a new light. This paper argues that Holland's fundamental arguments, and the years of developments that have followed, have a direct impact on systems of general network agents, regardless of whether they explicitly exploit EC. However, it also argues that the techniques and theories of EC cannot be directly transferred to the world of general (rather than EC-specific) agents, without examination of effects that are embodied in general software agents. This paper introduces a framework for EC interchanges between general-purpose software agents. Preliminary results are shown that illustrate the EC effects of asynchronous actions of agents within this framework. Building on this framework, coevolutionary agents that interact in a simulated producer/consumer economy are introduced. Using these preliminary results as illustrations, areas for future investigation of embodied EC software agents are discussed.