Cristian Gratie

Policy-Based Instantiation of Norms in MAS

Autonomy is a key feature in software agent design. For a system to function according to the expected behavior, however, the expectations must be expressed through a set of norms. The present work proposes a model for representing norms in multi-agent systems and for instantiating norms by using context-aware policies. The policies take into account both the general context and the reputation level of the target agent. The proposal aims at increasing the scalability and the interoperability of the normative systems implementing it, along with offering expressivity and richness in guiding agents towards the desired behavior.

Fuzzy labeling for argumentation frameworks

This paper introduces the use of fuzzy labels in argumentation. The first approach we propose is built as a natural extension of the in, out, undec labeling to real valued labels, coupled with an unsupervised learning algorithm that assigns consistent labels starting from a random initial assignment. The second approach regards argument (fuzzy) labels as degrees of certitude in the arguments acceptability. This translates into a system of equations that provides among its solutions the labelings that describe complete extensions.

General directionality and the local behavior of argumentation semantics

In abstract argumentation, the directionality principle conveys the intuition that, for an unattacked set, the choice of arguments that are part of an extension should only depend on the restriction of the framework to that set. Furthermore, having made such a choice, one should be able to select arguments from the rest of the framework so as to get an extension. In this paper we show how this idea can be generalized and used for formulating SCC-recursiveness as a stronger version of directionality. We argue that such properties characterize the information that is needed for computing the extensions of an argumentation semantics. We provide a formal approach for describing and comparing directionality-like properties. Our model provides a clear distinction between SCC-recursive semantics that use defense information and those that do not use it.

Agent-Based Information Sharing for Ambient Intelligence

In a future vision of Ambient Intelligence - or AmI - our surrounding environment will integrate a pervasive, interconnected network of devices. One important concern in AmI is the exact manner in which the large quantities of generated information are managed by a system formed mostly of devices with limited capabilities. This paper presents the first steps towards the realization of the AmIciTy framework: a multi-agent system that relies on local interaction and simple context measures, having as purpose the context-aware sharing of pieces of information. A prototype has been implemented and relevant simulation results are shown.

Alleviating Urban Traffic Congestion by Means of Adaptive Routing

Traffic congestion is an important problem of todays urban life. Various approaches have been analyzed for solving this problem, most solutions addressing only a specific cause of the congestion. This paper focuses on rush hour congestion and aims to address its cause, the fact that many drivers need to reach the same destination and use the same routes, most frequently the shortest ones.A simple solution would be a collaborative behavior from the drivers, so as to ensure that some of the drivers choose slightly longer routes in order to avoid congestion. This paper aims to investigate the improvement potential of using intelligent routing, while offering (as much as possible) the fastest available routes to the drivers.A simple yet effective routing algorithm is proposed and tested by means of a multi-agent system on top of a traffic simulator. Experimental results conform to the expected behavior of the algorithm and are meant to provide reference values for further research.

Emergent Properties for Data Distribution in a Cognitive MAS

Emergence is a key element in the research of multi-agent systems. Emergent properties provide higher level features of a system formed by simpler individuals. So far, emergence has been studied mostly by means of systems formed by reactive agents - that are more simple to design and implement. As computational power increases and even small devices become more capable, cognitive agents become a promising solution for the solution of more complex problems. This paper presents a simple example of a multi-agent system inspired from the Beliefs-Desires-Intentions model, designed to manifest emergent properties: a data storage system that assures distribution and replication of data without the need for these features to be controlled from the exterior.

Generic Representation for Extension-Based Semantics of Argumentation Frameworks

Argumentation is one of the most active domains in Artificial Intelligence, especially in multi-agent systems. Since the initial work of Dung on abstract argumentation frameworks, many semantics have been proposed, but for most practical cases argument acceptability is still a matter of choice. This paper introduces a formal model that aims to provide a unifying representation of argument extensions and thus offer a means for analyzing the relations between existing semantics. First, the generic model is described, then it is used for describing the most common argumentation semantics. Conclusions are drawn from the representation.

Full Hybrid μ-Calculus, Its Bisimulation Invariance and Application to Argumentation

Previous research has shown that argumentation semantics can be described with Monadic Second Order Logic. While certain less expressive, modal, logics can also capture some of the semantics, the general issue of finding minimal modal logics that are able to describe certain argumentation semantics has not received a lot of attention in the literature so far. In this paper we show that full hybrid μ-calculus cannot describe the preferred semantics, thus providing a negative answer to an open question. We show that the same holds for the skeptical and credulous versions of the preferred semantics. Our result relies on the invariance of full hybrid μ-calculus with respect to a suitable notion of bisimulation. We provide a complete proof of this invariance in the paper.