#52
You mean 'staff'.

Also, can I just repeat one more time that we're way off topic? :p

Aldo, any updates on this thing?
Not really. I've been working on my thesis proposal.

#53
You mean 'staff'.

Also, can I just repeat one more time that we're way off topic? :p

Aldo, any updates on this thing?
Not really. I've been working on my thesis proposal.
Oooh, fun.

#55
What do you plan to Aldo ? Hit me with the science!

1.2 Hypothesis
We hypothesise that:
Opportunistic repair strategies, that utilise a knowledge of functional dependencies, can provide robustness in Multi Agent Systems operating in hostile environments

Our research proposal is based on several key tenets. Firstly, we believe that the overall functionality of a multi-agent system can be defined in terms of the functional dependencies between agents forming that system. We define a functional dependency as existing where an agent can only perform a particular function (as indicated by possessing a particular capability) with the use of a function of another agent.

Should a particular agent suffer either a partial or complete loss of functionality (i.e. suffer debilitation), the overall impact upon the system can be determined through an understanding of which dependencies, if any, have been violated. A dependency is considered violated if an agent, depended upon to perform some particular task, is no longer capable of performing that task. This task may itself be dependent on certain performance or quality constraints, such as cost or temporal constraints; thus debilitation can violate a dependency without there being complete functional loss.

Secondly, we intend to underpin this understanding of dependency through a representation of capability; a definition indicating:
The functional tasks that can be performed by a particular agent.
The requirements of those tasks in terms of other capabilities.
The quality or performance offered by the agent possessing the capability.

This notion of capability is coupled with the dependency notion, conveying how the particular capabilities held by an agent may be dependent on usage of that agent's local actions, other agents' actions, or both.

Thirdly, and finally, we make an assertion that – given that the MAS can be understood in terms of functional interactions and dependencies – we can act to restore lost functionality through the application of repair strategies that act to compose behaviour through establishing new functional relationships and operational structures. These strategies will be opportunistic in nature, considering the existing state of the MAS to identify “opportunities” (available, suitable resources) and characterising the suitability of these based on the immediate needs of the repair operation.

Our proposed solution will be implemented as an extension of an existing agent framework, utilising a suitable and well-established simulation domain to implement a representative multi-agent system. This extension will be compared against the same MAS system, but with the repair mechanisms disabled, in order to evaluate the effect of agent failure in both systems.

Metrics will be defined based upon the generation and the execution of plans. We intend to prove that, when a repair strategy is implemented, the repair of functionality allows for generation of plans in situations where agent debilitation would otherwise prevent a suitable plan being formed, or where generated plans would offer only a suboptimal solution at best. The ultimate aim of this research is to produce a method that allows any MAS, structured on a task or goal decomposition basis, to be able to recover from agent debilitation which would otherwise have a severe 'domino effect' upon the larger scope of the MAS.

Who is online

Users browsing this forum: No registered users and 155 guests

cron