Logo Leibniz Universität Hannover
Logo: SRA/Leibniz Universität Hannover
Logo Leibniz Universität Hannover
Logo: SRA/Leibniz Universität Hannover
  • Zielgruppen
  • Suche

Current OC projects at SRA

Social OC: Trust in OC Systems

The first phase of OC research has targeted predominantly learning and self-optimizing single systems. In the context of Social OC we are now also interested in collective mechanisms important for the interaction of many agents. Key terms here are cooperation and competition, conflict resolution, altruism vs. egoism, the global detection of emergent behavior and the control of individual behavior through institutional mechanisms. As elucidated by game theory, trust is a key component of such systems, even when they are purely technical. Within the DFG Research Unit OC TRUST (in cooperation with the University of Augsburg) we investigate – using an open desktop grid computing system as an example -, how single agents behave in a trust-based infrastructure, and how the total system behavior can be kept within predefined boundaries by institutional mechanisms.

On-line Optimization

OC systems can be viewed as systems for distributed realtime optimization. This leads to a variety of research challenges: (1) OC agents permanently search a (virtual) fitness landscape but they act at the same time in this landscape thereby modifying it. Such a “self-referential” fitness landscape can be unpredictable and unstable. (2) Optimization must be fast enough such that its results are still meaningful. Therefore on-line optimization is not so much about finding a global optimum but rather about finding a feasible solution within limited time. (3) Search for an optimum is done concurrently by virtual search agents within the single subsystems and also by cooperation between (real) agents (subsystems).

We are interested in methods to speed up on-line optimization, among others by early search space pruning (tabu methods), by a-priori characterization of fitness landscapes, and by cooperative distributed optimization.

Parallel population-based Optimization on Multicores

Population-based optimization such as Particle Swarm Optimization (PSO) and derivatives are principally well suited for parallelization (e.g. on multi-cores). With the upcoming multi- and many-cores we will have computing resources available, which can also be used embedded into real systems. We investigate the distribution possibilities of population-based optimization algorithms on homogeneous and heterogeneous (GPGPU) architectures.

Application of OC e.g. in traffic control, robotics and communication networks

We use OC architectures and population-based optimization (GA, PSO etc.) to solve on-line optimization problems in application areas like 

  • Traffic control (throughput optimization, dynamic routing), 
  • Robotics (trajectory optimization, type synthesis – geometry and structure)
  • Communication networks (adaptation of routing parameters).