Rafał Dreżewski

Evolutionary Multi-Agent Systems

The aim of this paper is to give a survey on the development and applications of evolutionary multi-agent systems (EMAS). The paper starts with a general introduction describing the background, structure and behaviour of EMAS. EMAS application to solving global optimisation problems is presented in the next section along with its modification targeted at lowering the computation costs by early removing certain agents based on immunological inspirations. Subsequent sections deal with the elitist variant of EMAS aimed at solving multi-criteria optimisation problems, and the co-evolutionary one aimed at solving multi-modal optimisation problems. Each variation of EMAS is illustrated with selected experimental results.

A model of co-evolution in multi-agent system

Co-evolutionary techniques are aimed at overcoming limited adaptive capacity of evolutionary algorithms resulting from the loss of useful diversity of population. In this paper the idea of coevolutionary multi-agent system (CoEMAS) is introduced. In such a system two or more species of agents co-evolve in order to solve given problem. Also, the formal model of CoEMAS and the results from runs of CoEMAS applied to multi-modal function optimization are presented.

Co-Evolutionary Multi-Agent System for Portfolio Optimization

Co-evolutionary techniques for evolutionary algorithms can enhance the adaptive capabilities of evolutionary algorithms and help maintain population diversity. In this chapter the concept and a formal model of an agent-based realization of a predator-prey coevolutionary algorithm is presented. The resulting system is applied to the problem of effective portfolio building and is compared to classical multi-objective evolutionary algorithms.

The application of social network analysis algorithms in a system supporting money laundering detection

Criminal analysis is a very complex task requiring to process huge amounts of data coming from different sources such as billings and bank account transactions in order to gain knowledge useful for an investigator. In order to support human analytic capabilities, dedicated software tools are needed, and therefore Money Laundering Detection System (MLDS) was proposed as one of such tools in our previous paper. In this paper, the social network analysis component for this system is presented. The component makes it possible to use data from bank statements and the National Court Register and construct and analyze social networks during an investigation into money laundering cases. The system can assign roles to persons from the network and allows for analysis of connections between them. The paper also includes results of experiments aimed at investigating the performance of the implemented algorithms and the correctness of the analysis.

Multi-objective optimization using co-evolutionary multi-agent system with host-parasite mechanism

Co-evolutionary techniques for evolutionary algorithms are aimed at overcoming their limited adaptive capabilities and allow for the application of such algorithms to problems for which it is difficult or even impossible to formulate explicit fitness function. In this paper the idea of co-evolutionary multi-agent system with host-parasite mechanism for multi-objective optimization is introduced. In presented system the Pareto frontier is located by the population of agents as a result of co-evolutionary interactions between species. Also, results from runs of presented system against test functions are presented.

Multi-objective Optimization Technique Based on Co-evolutionary Interactions in Multi-agent System

Co-evolutionary techniques for evolutionary algorithms help overcoming limited adaptive capabilities of evolutionary algorithms, and maintaining population diversity. In this paper the idea and formal model of agent-based realization of predator-prey co-evolutionary algorithm is presented. The effect of using such approach is not only the location of Pareto frontier but also maintaining of useful population diversity. The presented system is compared to classical multi-objective evolutionary algorithms with the use of Kursawe test problem and the problem of effective portfolio building.

Classical and Agent-Based Evolutionary Algorithms for Investment Strategies Generation

In this chapter an evolutionary system for generating investment strategies is presented. The algorithms used in the system (evolutionary algorithm, co-evolutionary algorithm, and agent-based co-evolutionary algorithm) are verified and compared on the basis of the results coming from experiments carried out with the use of real-life stock data. The conclusions drawn from the results of experiments are such that co-evolutionary and agent-based co-evolutionary techniques better maintain population diversity and generate more general investment strategies than evolutionary algorithms.

Evolutionary system for generating investment strategies

The complexity of generating investment strategies problems makes it hard (or even impossible), in most cases, to use traditional techniques and to find the strict solution. In the paper the evolutionary system for generating investment strategies is presented. The algorithms used in the system (evolutionary algorithm, co-evolutionary algorithm, and agent-based co-evolutionary algorithm) are verified and compared on the basis of the results coming from experiments carried out with the use of real-life stock data.

A Co-evolutionary Multi-agent System for Multi-modal Function Optimization

Niching methods for evolutionary algorithms are used in order to locate all desired peaks of multi-modal landscape. Co-evolutionary techniques are aimed at overcoming limited adaptive capacity of evolutionary algorithms resulting from the loss of useful diversity of population. In this paper the idea of niching co-evolutionary multi-agent system (NCoEMAS) is introduced. In such a system the niche formation phenomena occurs within one of the preexisting species as a result of co-evolutionary interactions. Also, results from runs of NCoEMAS against Rastrigin function and the comparison to other niching techniques are presented.

An Agent-Based Co-Evolutionary Multi-Objective Algorithm for Portfolio Optimization

Algorithms based on the process of natural evolution are widely used to solve multi-objective optimization problems. In this paper we propose the agent-based co-evolutionary algorithm for multi-objective portfolio optimization. The proposed technique is compared experimentally to the genetic algorithm, co-evolutionary algorithm and a more classical approach—the trend-following algorithm. During the experiments historical data from the Warsaw Stock Exchange is used in order to assess the performance of the compared algorithms. Finally, we draw some conclusions from these experiments, showing the strong and weak points of all the techniques.