Courses
IFAAMAS Course - Introduction to Multiagent Systems (Michael Luck)
This course will provide an introduction to the field of agents
and multi-agent systems, spanning both the micro and macro
level, including: basic concepts; agent architecture;
interaction and multi-agent systems; applications; and future
challenges.
Logics for Multiagent Systems (John-Jules Ch. Meyer)
In this tutorial I will treat the basics of modal logic, and how
they are applied to describe agent attitudes and multi-agent
systems. Topics that will be addressed are basic modal logic,
epistemic logic, temporal logic, dynamic logic, deontic logic as
well as BDI logic, Common Knowledge and Joint Intentions, logic
as in applied in normative systems.
Agent Swarms Generating Short-term Forecasts and Increasing Situational Awareness (Paul Valckenaers)
In application areas such as logistics, traffic and
manufacturing, agent technology is employed to perform
coordination. This requires a look-ahead capability from the
agents. In a distributed setting, this look-ahead capability has
to account for interactions amongst the participants. Mainstream
negotiation-based agent technologies perform poorly in this
matter. Two designs based on swarms of light-weight agents offer
a solution.
The course compares two approaches for propagating the
intentions of situated agents through a virtual world, mirroring
some part of the real world. This propagation is executed by
swarms of lightweight agents. The propagation of agent
intentions causes short-term forecasts to emerge.
In a first approach, the world-of-interest is mirrored in a
“book” in which each page is a picture of the situation. Each
page corresponds to an instance in time, either the present,
recent past or imminent future. The state and the interactions
on page n determine the prediction of the system state on page
n+1.
In the second approach, the world-of-interest is mirrored in
a graph-shaped information structure. Each node representing a
resource (e.g. a road segment, truck, machine, etc.) has an
agenda. In this agenda, users make reservations that reflect
their intentions. The intelligent resources answer queries about
their future services whilst accounting for this agenda. This
second approach follows a travel arrangement pattern. Users
explore the graph-shaped world model through virtual execution
of a possible journey. Having explored sufficiently, the users
select a best-performing journey to become their intention, and
repeat the virtual execution of this journey whilst entering
reservations at the resources involved.
Trust and Reputation in Multiagent Systems (Guillaume Muller and
Laurent Vercouter)
The course is intended for people that want to get a wide
perspective on computational trust and reputations models.
The tutorial will be mainly composed of three parts. The first
part presents the general motivations and problems tackled by
trust and reputation models. It distinguishes these problems
from the complementary ones tackled by security techniques and
poses the foundations of trust models, based on sociological
works about trust and reputation.
The second part consists in a state of the art of the existing
works in the field of computational trust and reputation in
multiagent systems. It adresses prominent models proposed in
this domain: it starts from simple models like Marsh's trust
model or eBay and OnSale reputation model, then present more
decentralized models like Schillo and Funk's model and ends with
very rich and social models like ReGret and LIAR.
The last part focuses on the ART (Agent Reputation and Trust)
testbed. It is an open platform for experimentation and
competition for trust and reputation models. Its aims at
providing means to compare such models. The end of this last
part of the tutorial is dedicated to practice on the ART
platform.
Service Oriented Agents (Benjamin Hirsch)
One of the current hypes in the IT industry is the application
of service oriented architectures (SOA) to a wide range of
software systems that are distributed and often cross
organisational boundaries. It also appears that most of the
promises of SOA are also made by agent technology - be it
interoperability or agility - yet it is webservices and SOAP
instead of agents and speechacts that is en vogue today. It is
however difficult to discern between marketing hype and actual
contributions to the application of distributed systems.
The aim of this tutorial is to present the field of services and
service oriented architectures (SOA), and to show in how far
agents and SOA and related technologies can benefit from each
other. We believe that this topic is very relevant to aspiring
agent researchers as service orientation is of great importance
in the industry, and as the aims of agent research and SOA are
similar. We therefore want to create awareness of the
implications of SOA and the potential benefits the two areas can
have from each other.
The tutorial will consist of the following three parts:
Introduction to SOA: Students will be given a short introduction
to service orientation and SOA in general. The pros and cons of
SOA will be presented to, as well as practiced by the students
using handouts and small presentations.
How agents can contribute to Services and vice versa: In the
second part, we will focus on the common ground of agents and
services. We will point out different shortcomings of the
current SOA approach as well as possible solutions that agents
provide. Possible topics here are (formal) semantics in general,
ontologies, coordination and autonomous planning, goal oriented
behaviour and matchmaking. We also present possible lessons that
agent research can learn from SOA and associated technologies.
Current Approaches: The final part of the tutorial deals with
work at the interface between agents and services. In
particular, we plan to present our own agent framework JIAC, as
well as work to incorporate (web-)services, BPEL, and agents.
Introduction to Game Theory and Mechanism Design (Paul Harrenstein and
Matthijs de Weerdt)
In multi-agent systems multiple agents with individual preferences interact in a
common environment. For the design and proper formal understanding of the
interaction in multi-agent systems appropriate mathematical concepts are
required. Game theory provides and analyzes such concepts as equilibrium
outcomes and strategies.
Roughly speaking, mechanism design is concerned with the development of
multi-agent systems with desirable game-theoretic properties even if the
preferences of the agents are unknown to the designer or to the other agents. It
is also concerned with what is formally possible and impossible in this respect.
Put this way, the relevance of game theory and mechanism design for multi-agent
research might seem obvious. Nevertheless, game theory nor mechanism design are
commonly part of a standard curriculum in computer science. This course aims to
fill this gap.
This introductory course is aimed at postgraduate students in computer science
and multi-agent systems who have no previous acquaintance with game theory or
mechanism design. As such only some elementary mathematical training is assumed.
The main goal of the course is to make the students familiar with the most
fundamental concepts and results of game theory and mechanism design and to make
clear how game theory and mechanism design can be applied to their own research.
Wireless Sensor Networks and Multiagent Systems (Eric Platon and
Danny Weyns)
Wireless sensor networks (WSN) are networks of small computers endowed with
sensors and wireless connectivity. The comparison of WSN with multi-agent
systems (MAS) is natural, as the two approaches consist of interactive entities
situated in an environment they can sense and act upon locally. Agents provide
engineers with a higher level of abstraction, so that WSN become a useful
application domain for MAS. MAS researchers have actually developed adequate
models and approaches to address some current research issues in the area of WSN
(e.g. self-adaptation and collective behaviors).
The goals of this tutorial are to (i) introduce WSN technologies, (ii) show why
MAS are important for the development of WSN, (iii) explain the main research
challenges in WSN, and (iv) show how MAS can address relevant issues.
This tutorial is a half-day session tailored for researchers and practitioners
interested in the application of MAS to WSN. It assumes background knowledge in
MAS and basics of networks, although it will summarize important information
when necessary.
Related events: Workshop on Agent Technologies for Sensor Networks (AAMAS);
International Conference on Networked Sensing Systems; Conference on Embedded
Networked Sensor Systems; European Conference on Wireless Sensor Networks,
RoboCup.
Foundations of Institutions (Marina De Vos,
Marc Esteva and
Julian Padget)
Institutions are multi-agent systems in which agents are
governed by norms. Agents can break the norms provided that they
are willing to accept the ensuing sanction. With agents not
constantly abiding by the rules, it becomes important that
institutions can monitor and penalise any violations that occur.
In this six hour tutorial we will: - Discuss the basic
principles of institutions and their components; - Present a
general framework that allows a user to specify, design,
reasoning about and verify institutional models; - Describe and
demonstrate graphical notations and tools for the creation and
animation of institutions.
The primary delivery method will be by lecture alleviated by
class discussion and exercises.
Agents and Arguments (Sanjay Modgil and
Leila Amgoud)
An argument is a set of premises offered in support of a conclusion, and
argumentation is the process whereby arguments are constructed, exchanged and
evaluated in light of their interactions with other arguments. Logical models of
argumentation have recently emerged as a promising paradigm for modeling agent
reasoning and communication. The paradigm's promise resides in its modular
nature that closely mirrors the way humans reason. It thus provides a general
framework for inference and decision making over the full gamut of mental
attitudes (beliefs, goals, desires, actions e.t.c) in the presence of
inconsistent, uncertain and incomplete information. The generality of the
argumentation paradigm is substantiated by the fact that existing logical
approaches to non-monotonic and commonsense reasoning can be formalised in terms
of argumentation. Furthermore, argumentation has recently been used to enrich
models of communication. It provides a principled way in which to structure
rational dialogue and enable exchange of, and reasoning about,
justifications/arguments for proposals and or statements between human and or
automated agents. This course aims to introduce students and researchers to the
fundamentals of logic based models of argumentation, and to present recent
research work on development of these models for application in agent reasoning
and communication. We review: 1) Abstract argumentation frameworks and
extensions to these frameworks to facilitate flexible and adaptive agent
reasoning. 2) Instantiations of these abstract frameworks for formalising agent
inference and decision making over mental attitudes. 3) Frameworks for
argumentation based dialogue, whereby agents argue in order to persuade or
negotiate with other agents.
Computational Complexity in Multiagent Systems (Edith Elkind and
Evangelos Markakis)
In many multi-agent settings, agents have limited computational
resources, and cannot function in environments where each
decision requires large amounts of computation. Therefore,
computational complexity is an important parameter in designing
and evaluating multiagent systems. It provides a useful tool to
the agent designer, enabling him to choose whether to focus on
finding an optimal strategy, or use approximate or heuristic
approaches.
In this tutorial, we cover foundations on computational
complexity and its applications in the context of multiagent
systems. We first introduce some of the most important
complexity classes, such as P, NP, PSPACE and #P, as well as the
basic notions of computational complexity, such as reductions,
and explain what it means for a problem to be hard or complete
for a given class. We then provide examples of multiagent
decision problems that are hard for the above mentioned classes,
and discuss the implications of these results from a practical
perspective. Usually, a hardness result motivates research in
heuristic and approximation algorithms for the problem, and we
provide several examples where such approaches succeed. On the
other hand, we also demonstrate that computational hardness can
be used as a barrier to undesirable behavior by strategic
agents, in settings where a truthful agent's decision problem is
easy, while finding a way to manipulate the system to an agent's
benefit is computationally hard. Our examples will be drawn from
various areas of research in multiagent systems, such as
multiagent planning, argumentation, auctions, coalition
formation and voting, illustrating the importance of
computational complexity concepts in these domains.
The course is intended for 1st-year PhD and MSc students with
little or no background in computational complexity.
Planning in Multiagent Systems (Matthijs de Weerdt and
Cees Witteveen)
By definition, agents are autonomous entities that are able
to act. Hence, the process of determining which actions to
execute and the order in which these actions have to be executed
is considered as an essential property of agents. Within the
AI-community such processes have been studied in the context of
planning problems.
Traditionally, the planning community concentrated on
planning as a single-agent phenomenon, focusing on aspects of
representation, efficiency and robustness of a plan executed by
a single agent. In a multi-agent context, however, other aspects
of the planning process become more important. Nowadays a major
issue in multi-agent planning is the coordination of cooperative
as well as self-interested agents. The focus on coordination in
planning also brings in other major topics in multi-agent
systems like auctions, coalition formation, privacy,
scalability, etc.
In this tutorial we start with an overview of multi-agent
planning problems and discuss the properties of these problems
that have the most influence on the applicability of certain
existing approaches. We will give an overview of these
approaches by distinguishing their role in the planning and
coordination phases of a multi-agent system. We then discuss how
single agent AI planning techniques lead to different
multi-agent planning approaches. Finally, we discuss some
particular approaches to coordination of single-agent planning
systems in more detail.
What Coalitions Can Achieve (Juergen Dix and
Wojtek Jamroga)
In this course, we present some formal approaches to modeling and reasoning
about strategic abilities of groups of agents. The course consists of two parts.
The first part addresses prescriptive concepts (mostly from collaborative game
theory) that specify how much a coallaboration between agents is worth and what
it means for a team to be "good". We also discuss various search algorithms to
find such good coalitions (or, ones within a certain bound from the optimal
coalition). The focus of the second part is descriptive. We introduce a modal
logic of strategic ability, ATL, and show how abilities of coalitions can be
specified in that language. Finally, we discuss the algorithmic side of checking
such specifications. The course requires some elementary knowledge of logic and
game theory. Familiarity with basic modal logic will be an advantage.
Preliminary outline of the course:
1. How to Form a Coalition. - Basic models & concepts (brief overview):
strategic games, extensive games, from extensive to strategic form, games with
imperfect information; Nash equilibrium, Pareto optimality. - Coalition
formation: definition of the problem, core, CS-search, payoff division (Shapley
value). - Contract Nets.
2. Reasoning about What Teams Can Achieve. - Modal logic (brief intro): models,
operators, epistemic logic. - Reasoning about abilities of agents and teams:
alternating-time temporal logic ATL for perfect information games, problems with
imperfect information, ATLir. Model checking abilities of coalitions. -
Reasoning about rational teams.
Course webpage:
http://www2.in.tu-clausthal.de/~wjamroga/courses/Coalitions2008EASSS/
Agent Oriented Software Engineering (Onn Shehory)
Agent Oriented Software Engineering (AOSE) is a key factor for
introducing agent-based systems to the industry as an
engineering approach. At present, the majority of existing agent
applications are developed in an ad hoc fashion: little or no
rigorous design methodology, limited specification of the
requirements, ad hoc design of agents and of multi-agent systems
as a whole, and little attention to non-functional requirements
such as mobility, scalability, performance issues, and
standards. By adopting AOSE principles, one gains the advantages
of an organized development process such as reusability,
testing, and maintenance. One of the basic principles of AOSE is
using a methodology for developing agent applications. Hence,
this course will concentrate on methodologies, their
applicability, and their use. In particular, the goals of this
course are the following: -Introduce basic concepts of software
engineering in the context of agent-based systems; -Introduce
the motivation for using agent-oriented software engineering;
-Introduce the field of agent-oriented methodologies; -Present
several agent-oriented methodologies; -Compare exiting
agent-oriented methodologies; -Discuss implementation issues of
agents and MAS and their relationship to agent-oriented
methodologies.
Automated Negotiations in Electronic Markets (Nicola Gatti)
Agents are well-suited for dynamic, constrained, and real-time
environments such as electronic marketplaces. In such
environments agents representing their users negotiate for goods
and services following negotiation protocols. Bargaining and
auctions are the principal negotiation protocols for buying and
selling goods based upon competition among the interested
parties. This tutorial will introduce participants to
agent-based negotiations.
The tutorial will start by introducing agent-based negotiation
and negotiation protocols in general. Bargaining and auctions
will then be described in detail. Essential concepts that are
required for following the tutorial will be introduced along the
way. The bargaining problem will be introduced and the principal
non-cooperative bargaining protocol (Rubinstein's
alternating-offers) will be discussed in detail. Variations and
applications Rubinstein's protocol in computer science will be
presented. The four single side auction protocols (English,
Dutch, FPSB, Vickrey) will be discussed in depth along with
their relative advantages and disadvantages. Double auctions and
the M-th and (M+1)-st clearing rules will also be covered. A
brief exposition into more advanced auction formats such as
multi-attribute and combinatorial auctions will follow.
Agent Based Simulation for Social Studies (Luis Antunes and
Federico Cecconi)
Agent-based social simulation is a recent multi-disciplinary
effort that has increasingly established new challenges for the
agents community, by bringing the agent technology to face
complex phenomena such as the ones found in social sciences.
At the same time, social scientists have been discovering how
the computer and especially the advances in the agent field can
provide a new and exciting tool to tackle the problems of their
field, providing a paradigm shift in social sciences. The
exchange between researchers in both areas has proven mutually
fruitful, as much inspiration in MAS has come from social
sciences.
The course begins with an introduction to multi agent modelling:
the main issues are the micro-macro links and the concept of
rationally-heterogeneous agents. Then, the tutorial will deal
with the main concepts of game theory and dynamical systems,
following this outline: an overview of game theory and
equilibrium (Nash, repeated games, Bayesian games); learning and
evolutionary games (ESS); dynamical systems; Monte-Carlo and
numerical simulations. Finally, agent based methodology will be
discussed: (a) the problem of the level of description, (b) the
relationships with mathematical modelling, (c) the tools and the
problem of scalability.
Using NetLogo, students will carry out some experiments: (a)
cooperation in minority games; (b) reputation's dynamic; (c)
social capital in a social networks; (d) imitation vs. evolution
in an artificial trade market.
Normative Multiagent Systems (Guido Boella,
Davide Grossi and
Leon van der Torre)
The Agentlink Roadmap, published in 2005, considers norms as key
for the development of MAS. Applications of norms range from
agent organizations and electronic institutions (e-commerce, and
e-government), to open agent societies, agent communication,
trust and reputation systems, and MAS programming. Norms are
among the social notions which are obtaining most attention in
the MAS research area (see, for instance, the COIN and NorMAS
workshops, the last editions of the DEON conference), bringing
together MAS researchers, legal and social scientists,
logicians.
The tutorial provides a clear understanding about when norms
become important in designing and developing MAS, about what
kind of norms are of importance for MAS and how they should be
used. A student session is planned on the application of norms
to the students' research projects. The tutorial is given by
leading scientists in the field of Normative Multi-Agent
Systems, organizers of the NorMAS'05-'07-'08 workshops and of
the DEON'08 conference. The tutorial is structured in a session of three lectures and a
student session: 1. When are norms to be used?: The First part
of the tutorial provides an introduction to Normative
Multi-Agent Systems stressing what are the MAS-related topics in
which norms play a key role; 2. What kind of norms are to be
used?: The second part of the tutorial focuses on the different
types of norms (and their interaction) which are of particular
relevance for the design of Normative Multi-Agent Systems; 3.
How are norms to be used?: The third part of the tutorial shows
how norms can be used to support the design of several aspects
of Normative Multi-Agent Systems and how norms relate to other
social notions such as roles, organizations and institutions; 4.
Student session: Students will be actively involved by
discussing relations of their research questions to the topics
dealt with in the tutorial. |