Abstract

This work presents our contribution about the distribution of complex system simulations in open distributed environments. We consider these environments as computational ecosystems, of whose we describe properties and characteristics, wherein simulations will evolve, by their execution. An interaction network models these simulations, that we represent as a dynamic graph. Considering the different kinds of possible dynamics, we propose a global formalism that can be used to describe these graphs, along with a software framework, GraphStream, allowing their manipulation and their study.

The dynamic graph is then the object of a study wherein we are looking for organisations, following them through time, in order to minimise communications’s costs between machines, and to balance the computation load. We bring an improvement aiming to reduce results’s oscillations of AntCo 2 , which is algorithm using digital ants colonies to detect organisations, through competition and collaboration mechanisms. Stability of these organisations is determined by a dynamic and distributed heuristic to find centroid of graph.

We propose a middleware allowing a dynamic and decentralized distribution of simulations in a computational ecosystem, by favouring organisations and respecting the load balancing.

Abstract

There are many segmentation methods in image processing. These methods are usually limited to a few image types. To our knowledge, there is no generic method which is able to solve segmentation problem for any image. In this master thesis, a segmentation method which uses a socials spiders colonies system will be studied and improved. These spiders are able to collectively weave silk structure. This master thesis aims to present an adaptation of this method to the segmentation problem.

[ˆ1] authors are sorted alphabetically