Solving Distributed Dynamic Optimization Problems in Self-optimizing Systems by Approximating the Interaction between Agents

Abstract:
This paper presents first approach towards the approximation of the interaction between agents in order to solve distributed dynamic optimization problems in Organic and Autonomic Computing systems. Existing Organic Computing systems based on multiple agents often develop towards a non-optimal solution due to a greedy local optimization of each agent. Obtaining the optimal solution requires to impair the local state of some agents. Therefore, the proposed approach approximates the interaction between agents by predicting how the actions of an agent influence their neighboring agents. The goal is to only choose actions that benefit the community of agents instead of greedy,selfish h actions. Exemplarily, a dynamic optimization problem is considered to quantify the effectiveness of the discussed procedure.