Ruby L. V. Moritz

Refined ranking relations for selection of solutions in multi objective metaheuristics

Two methods for ranking of solutions of multi objective optimization problems are proposed in this paper. The methods can be used, e.g. by metaheuristics to select good solutions from a set of non dominated solutions. They are suitable for population based metaheuristics to limit the size of the population. It is shown theoretically that the ranking methods possess some interesting properties for such applications. In particular, it is shown that both methods form a total preorder and are both refinements of the Pareto dominance relation. An experimental investigation for a multi objective flow shop problem shows that the use of the new ranking methods in a Population-based Ant Colony Optimization algorithm and in a genetic algorithm leads to good results when compared to other methods.

Evolutionary Inheritance Mechanisms for Multi-criteriaDecision Making in Multi-agent Systems

In this paper we study the use of different evolutionary inheritance mechanisms for the adaptation of parameters in a multi-agent system where the agents have to solve tasks that are distributed within a dynamic environment. In the studied system the agents have to form teams to execute the tasks. Deciding which task to execute next is a multi-criteria decision problem for which the agents use different ranking schemes. Agents that have successfully executed several tasks can reproduce and pass the type of ranking scheme they have used and some corresponding parameter values to their successors. Three types of evolutionary mechanisms are compared: haploid, diploid, and haplo-diploid. The latter one is new for multi-agent systems. The focus of our simulation experiments is to study the influence of the different evolutionary mechanisms on the diversity of the agents and on the resulting efficiency of the multi-agent system for different dynamic environments.

The Influence of Correlated Objectives on Different Types of P-ACO Algorithms

The influence of correlated objectives on different types of P-ACO algorithms for solutions of multi objective optimization problems is investigated. Therefore, a simple method to create multi objective optimization problems with correlated objectives is proposed. Theoretical results show how certain correlations between the objectives can be obtained. The method is applied to the Traveling Salesperson problem. The influence of the correlation type and strength on the optimization behavior of different P-ACO algorithms is analyzed empirically. A particular focus is given on P-ACOs with ranking methods.

Self-organized cooperation between agents that have to solve resource collection tasks

We analyze the behavior and efficiency of a task-force of heterogeneous agents, which have different skills to solve a resource collection task. The agents can move within an arena and are able to cooperate in order to accessing the skills of other agents. Cooperation between agents is only possible, when they form a group by moving to the same location. Yet the number of agents on the same location is restricted. Thus the agents can only cooperate with a limited number of other agents at a time. Different strategies for the agents to decide whether to cooperate or not and how to detect the most favorable partner for cooperation within their surrounding are are investigated. In the studied scenarios the amounts of the different resources that are available varies between different locations. The agents are forced to decide whether to follow the drive of their own efficiency, or rather to build a cooperation with close-by agents that might be located in a different direction.

Decentralized and dynamic group formation of reconfigurable agents

Adaptive group formation in dynamic environments performed by heterogeneous swarms of simple agents is an interesting research topic. In this paper we consider an unsupervised scenario where the individuals of the swarm have limited information about their environment as well as limited communication capabilities. The particular case of a multi-agent model with self-organized reconfigurable agents where the agents are confronted with a resource collection task, different movement, and group formation tactics are analyzed experimentally. It is shown that cooperation in groups is profitable for the group members and the optimal group size depends on environmental parameters. Moreover, a simple strategy based on the agents ability to measure their own workload results in an adaptive behavior that influences the size of the groups and increases the performance of the overall system.

Self-adaptable Group Formation of Reconfigurable Agents in Dynamic Environments

The formation of groups in heterogeneous swarms is important whenever there is benefit from cooperation between the members of a group. Here we investigate decentralized group formation strategies for a set of moving agents that have to collect resources. The agents are reconfigurable and can adapt themselves to the needs of their group. It is assumed that the agents are simple and communication between groups is limited to sensing each other. Members of one group are able to share information about their capabilities and movement decision. Also the decision strategies for integrating a new agent into a group and for the moving direction of a group are simple. Several versions of the system are compared experimentally for a dynamic situation where the number of available resource types changes. It is shown that the costs for reconfiguration influences the optimal strategy for the integration of new agents into a group. The system can adapt the average group size to the number of available resource types.

Annotation guided local similarity search in multiple sequences and its application to mitochondrial genomes

Given a set of nucleotide sequences and corresponding gene annotations which might contain a moderate number of errors we consider the problem to identify common substrings occurring in homologous genes and to identify putative errors in the given annotations. The problem is solved by identifying nodes in a suffix tree that contains all substrings occurring in the data set. Due to the large size of the targeted data set our approach employs a truncated version of suffix trees. The approach is successfully applied to the mitochondrial nucleotide sequences and the corresponding annotations available in RefSeq for more than 2000 metazoan species. We demonstrate that the approach finds appropriate subsequences despite of errors in the given annotations. Moreover, it identifies several hundred errors within the RefSeq annotations.

Trophallaxis-inspired self-organized task exchange in heterogeneous swarms

A trophallaxis-inspired method for self-organized task exchange in a swarm of autonomous, movable, and recon-figurable agents is proposed. Two agents that meet each other can exchange tasks with each other such that each one receives tasks that fit better to its current configuration. This reduces the energy needed for reconfiguration in order to execute the received tasks. It is investigated how an increase in the speed of moving of an agent can increase its chances for getting better tasks because of a higher frequency of meetings with other agents. An interesting trade-off is studied between getting better tasks when moving faster but losing more energy for the movement. Static scenarios where the agents spent a constant fraction of their energy on movement as well as dynamic scenarios are considered and evaluated by extensive simulations.

Refined ranking relations for multi objective optimization andapplication to P-ACO

Two new ranking methods for solutions of multi objective optimization problems are proposed in this paper. Theoretical results show that both new ranking methods form a total preorder and are refinements of the pareto dominance relation. These properties make the ranking methods suitable for the selection of a subset of good solutions from a set of non-dominated solutions as needed by meta-heuristics. In particular, this is shown experimentally for a Population-based ACO that uses the ranking methods to solve a multi objective flow shop problem.