Adaptive organisation centered multi-agent systems

• Autores: Jordi Campos Miralles
• Directores de la Tesis: Maite López Sánchez (dir. tes.) , Marc Esteva Vivancos (dir. tes.)
• Lectura: En la Universitat de Barcelona ( España ) en 2011
• Idioma: inglés
• Tribunal Calificador de la Tesis: Josep Lluís Arcos Rosell (presid.) , Inmaculada Rodríguez Santiago (secret.) , Alexander Artikis (voc.)
• Texto completo no disponible (Saber más ...)
• Resumen

  • Adaptive Organisation Centered Multi-Agent Systems by Jordi Campos This thesis focuses on the design and development of Adaptive Organisation Centred Multi-Agent Systems (AOCMAS) in non-task-oriented regulated scenarios. In short, Multi-Agent Systems (MAS) are computational systems where a set of autonomous software components (agents) interact within an environment. Within MAS, the Organisation Centred Multi-Agent Systems (OCMAS) have proven to be successful in promoting a coordination model that structures agent interactions. However, changes in agents' behaviour or in the environment may lead to a poor fulfilment of the system goals, and so, its entire organisation needs to be adapted. We refer as Adaptive OCMAS (AOCMAS) to those systems that are able to dynamically adapt their organisational components to better accomplish their goals. In particular, we advocate for endowing the organisation with adaptation capabilities, instead of expecting agents to be capable of adapting the organisation by themselves. Furthermore, we propose a solution based on a regulative approach instead of subtasks. Hence, our solution is able to deal with problems in which there is no goal that can be decomposed into tasks assigned to agents ---e.g. improving traffic flow in a road-network.

    We argue that existing OCMAS' components are devoted to enact the coordination model mentioned above. Hence, we review the existing OCMAS works as a set of coordination support mechanisms. Even more, from this point of view we envision a set of mechanisms that assist coordination further than just enacting it. In particular, we regard the adaptation of an organisation as an assistance service that provides an added value to merely enabling the organisation existence. For this purpose, we present a formalisation of such an adaptation service and an abstract architecture to implement it, the so-called 2 LAMA.

    We exemplify all these concepts in different problems driven by goals. Nonetheless, we illustrate our whole approach by applying it to a case study in a Peer-to-Peer sharing network (P2P) scenario. In particular, we analyse the adaptation of agents' social structure and a set of norms ---both of them are organisational components. Moreover, due to the complexity of such processes, we study two alternatives to perform the latter. That is, we adapt norms by using a heuristic approach and a machine learning technique. Specifically, we use an extension to Case-Based Reasoning machine learning technique that is able to deal with multidimensional continuous search spaces, unknown optimal solutions and awkwardness to identify the impact of a solution on the final result.

    A relevant effort of this thesis has been to develop a simulator that enables to compare different coordination models in the P2P case study. It provides several tools to extract and analyse measures from communication network level up to agent level. Furthermore, it has a modular mechanism to easily test different norm adaptation approaches. We use it to empirically evaluate our proposals. The results show that the cost of introducing an additional layer in charge of the system's adaptation is lower than its benefits.