|
The IP Proceedings
Abstracts:
Title:
English version:
Various nature of knowledge and their representation
Abstract:
Agents, as every intelligent system, need some kind of knowledge.
It can eventually be represented as pure procedures (compiled knowledge),
but it is more useful, especially for the interaction with humans,
to have it under more declarative forms. The basic requirement is
to be able to draw conclusions from the input, i.e. to perform inferences.
Therefore, any representation of knowledge can be seen as a kind
of Logic. But there exist several kinds of Logic, and this lecture
will try to draw correspondences between the intrinsic variety of
knowledge and the set of existing deductive systems. After a short
reminder concerning propositional, first-order, higher-order classical
logics, and many-valued logics, a brief presentation of modal and
non-monotonic logics will be given, in relation with semantic networks,
and in each case, the problem of adequacy of the nature of knowledge
with the logic considered will be raised.
French version: Les
différentes natures de connaissances et leur représentation
Abstract:
Les agents, comme tout système intelligent, ont besoin de se servir
de connaissances. En fin de compte, elles peuvent être représentées
comme de pures procédures (connaissances compilées), mais il est
plus utile, surtout en vue de l'interaction des agents avec les
êtres humains, de disposer de ces connaissances sous des formes
plus déclaratives. L'exigence de base est la possibilité de tirer
des conclusions à partir de données, c'est-à-dire d'effectuer des
inférences. C'est pourquoi toute représentation de connaissances
peut être vue comme une sorte de logique. Mais il y a de nombreuses
sortes de logique, et cet exposé tentera d'établir des correspondances
entre la diversité intrinsèque des connaissances et l'ensemble des
systèmes déductifs existants. Après un bref rappel sur les logiques
classiques (propositionnelles, de premier ordre, d'ordre supérieur)
et sur les logiques multi-valuées, on présentera rapidement des
logiques modales et non monotones, en relation avec les réseaux
sémantiques, et dans chaque cas, l'adéquation de la nature des connaissances
avec la logique considérée sera discutée.
Author: Prof. Daniel
Kayser
Title:
English version:
Multi-agent Modal Logics
Abstract:
The lecture will focus on the work of Halpern, Moses & al. Generally,
modal logic considers one relation of accessibility. What happens
if several such relations are simultaneously present? This is an
abstraction of a multi-agent system where each agent is characterized
by its own relation of accessibility. Notions such as common knowledge,
implicit knowledge, etc. are defined and their properties are developped.
French version: Logiques
modales pour les systèmes multi-agents
Abstract:
L'exposé portera sur les travaux d'Halpern, Moses & al. Généralement,
la logique modale considère une seule relation d'accessibilité.
Qu'arrive- t-il si plusieurs relations sont simultanément présentes
? On obtient ainsi une abstraction d'un système multi-agent, dans
laquelle chaque agent est caractérisé par sa propre relation d'accessibilité.
Des notions telles que les connaissances communes, les connaissances
implicites, etc. seront définies et leurs propriétés seront développées.
Author: Prof. Daniel
Kayser
Title:
A unified model
of negotiation based on argumentation
Abstract:
The paper presents a multi-agent system that comprises a society
of self-interested agents that use argumentation-based negotiation
to reach agreements regarding cooperation and goal satisfaction.
The system is a generalization of some argumentation-based multi-agent
systems proposed in the literature in which better cooperation agreements
are reached through the use of human-like arguments. We then show
how this type of negotiation can be adapted according to evolved
models of other agents in the system. Negotiation is performed using
different types of arguments varying from quantitative ones, such
as money or trade objects, to qualitative arguments, such as promises,
appeal to past promises, and past examples. The models of the other
agents are built and refined incrementally during negotiation; these
models are then used to adapt the negotiation strategy according
to other agents' desires, preferences and behavioral characteristics
during interactions.
Author: Cosmin
Carabelea
Title:
English version: Learning
in MAS
Abstract:
Machine Learning can be defined as the explicit acquisition of new
knowledge in order to improve the performance of a system. The lecture
will briefly review the definitions and some characteristics of
Machine Learning in Artificial Intelligence. Then Machine Learning
will be studied along the four dimensions of MAS : Agents, Environment,
Interactions, and Organisation. In each dimension, some examples
of learning MAS will be presented. Then we shall try to answer some
basic questions such as : "Why to learn in a MAS ?" ; "What can
be learned in a MAS ?" ; "How to learn in MAS ?". In conclusion
we shall present some works about Assistant Agents and Machine Learning.
French version: Apprentissage
dans les SMA
Abstract:
L'apprentissage automatique peut-être défini comme l'acquisition
d'une connaissance explicite de façon à améliorer les performances
d'un système. L'exposé passera rapidement en revue quelques définitions
et caractéristiques de l'apprentissage automatique en intelligence
artificielle. Ensuite nous étudierons l'apprentissage automatique
dans les SMA selon les quatre dimensions d'un SMA : Agents, Environnement,
Interactions et Organisation. Dans chaque dimension des exemples
de SMA apprenants seront présentés. Ensuite nous tenterons de répondre
à quelques questions de base telles que "Pourquoi apprendre dans
un SMA ?" ; "Quoi apprendre dans un SMA ?" ; "Comment apprendre
dans un SMA ?". En conclusion nous présenterons quelques applications
de l'apprentissage automatique aux agents d'interface.
Author: Prof. Philippe
Beaune
Title:
Protocols for
negotiation with coalition between self-interested agents
Abstract:
Negotiation is a kind of decision making where two or more parties
jointly search a space of solutions with the goal of reaching a
consensus. This search is usually a single dimensional process where
the different agents involved on the negotiation adapt their price
offers in order to reach their goals (to sell or to buy a good).
However, this is a limited process many times unrealistic because
more than one single issue must be negotiated in order to find an
adequate solution. In this talk we present a multi-issue negotiation
protocol for one buyer to many sellers interaction. The presentation
is enriched by an application example that illustrates the negotiation
process and demonstrates how it can be useful both as a product
brokering and a market brokering tool.
Keywords:
Negotiation, Multi-issue negotiation, Electronic Commerce.
Author: Prof. Jose
Ribeira da Fonseca
Title:
Models of communication
for multiagent systems
Abstract:
Interaction between autonomous agents within a multiagent system
requires explicit communication in order to allow knowledge exchange
between them. This seminar deals with the interaction issue between
agents from different perspectives. First, one will discuss interaction
languages among agents and will show how they apply on different
domains. Second, one will focus on interaction protocol engineering
and tackle the modeling of a structured message exchange between
agents according to a component-based approach. A related application
such as electronic commerce will be presented. Thirdly, one will
present the notion of constraints on interaction protocols via a
calculus model called POS which is based on operationnal semantics.
Such an approach will be shown upon coordination strategies for
soccer-robots.
Author: Prof. Jean-Luc
Koning
|
