Tutorials

Monday, 22 November 2004

If you have any question about tutorials or if you have already paid the corresponding tutorial fees,
please contact the Tutorial Chair, Dr. Angelica Munoz


Rotor: Microsoft shared source CLI implementation
Armando Halbinger, Microsoft - México

The Microsoft Shared Source Code CLI implementation, also know as “Rotor”, is a complete implementation of the ECMA-334 (C#) and ECMA-335 (CLI) standards in source code form. In its million-plus lines of source code, you will find compilers, tools, techniques for automatically managing memory, just-in-time (JIT) code generators, component and Web services infrastructure, globalization know-how, security protocols, and all sorts of other intriguing realizations of abstract concepts.

Outline:
   Introduction to CLI component model
   CLI type system
   Assemblies: component packing
   JIT compilation
   Managed code execution engine
   Garbage collection
   Research Projects using Rotor

About the speaker: Armando Halbinger is the Academic Developer Evangelist at Microsoft Mexico. Previously, he was a Developer Consultant for 3 years at Microsoft Enterprise Services. He holds a bachelor’s degree in Computer Science from the ITESM-CCM in Mexico City. His areas of expertise are .NET development, software architectures and software processes.

From planning to multi-agent planning: issues, models and case studies (4 hours/English)
Amal El Fallah Seghrouchni, University of Paris X, LIP6 (University Pierre and Marie Curie and CNRS-UMR 7606) - France

One of the major interests of Multi-Agent Systems (MAS) is their ability to handle distributed planning by coordinating agents' plans. Coordination requires both an adequate plan representation and an efficient interaction between agents. Based on information exchange (e.g. data, plans), the interaction allows agents to update their own plans by considering the exchanged information. Coordination generally produces two effects: cancelling negative interactions (e.g. harmful actions) and taking advantage of helpful interactions (e.g. handling redundant actions). Agents organize their activities and update their plans in order to cooperate and avoid conflicts.

Even if the planning topic has been well-addressed, the topic of multi-agent planning (MAP) is less known since it is situated at the junction of two domains: MAS and Planning. MAP is an important mechanism for MAS coordination. From the theoretical point of view it raises several interesting issues because of the main features of agents (autonomy and partial view of agents) and the environment changes. MAP should be studied from several perspectives: dynamic (or reactive) planning, distributed planning, task allocation and resource sharing, etc.

Tutorial Content: 
   A brief overview on non-classical planning approaches (in the case of single agent planning) such as: non linear and temporal planning.
   A classification of the main MAP approaches, in particular distributed versus centralised approaches, task-oriented versus agent oriented planning.
   Two approaches:  reactive MAP and temporal MAP; will be illustrated through some significant case studies such as aircraft planning, and robotics.
   The necessary background and potential target audience for the tutorial:
   Classical planning (basic notions on single-agent planning)

A knowledge of  multi-agent systems should be useful (but not necessary)

About the speaker: Amal El Fallah Seghrouchni is Professor at the University of Paris X. She is also searcher at LIP6 laboratory (University Pierre and Marie Curie and CNRS - UMR7606). Her main research topics are "Analysis, Design and Validation of Multi-Agent Systems". She published more than 60 papers in the major conferences of MAS field, co-edited 3 books and co-organised several international events dedicated to Multi-Agent Systems. She also has been invited professor and gave talks in international seminar and courses for up-graduated students about multi-agent planning, coordination, interaction protocols, etc. (For more details, please see: www-poleia.lip6.fr/~elfallah).

Automatic text summarization: past, present, and future (8 hours/English and/or Spanish)
Horacio Saggion, University of Sheffield - UK

A summary is a condensed version of a source document having a recognisable genre and a very specific purpose: to give the reader an exact and concise idea of the contents of the source. In most cases, summaries are written by humans, but nowadays, the overwhelming quantity of information and the need to access the essential content of documents accurately in order to satisfy users' demands calls for the development of computer programs able to produce text summaries. This is a difficult task because it requires linguistic and world knowledge which are difficult to incorporate in automatic systems. Even though some approaches to text summarization produce acceptable summaries for specific tasks, it is generally agreed that the problem of coherent selection and expression of information in text summarization is far from being solved, thus making  it an interesting research topic as demostrated by the TIDES (Translingual Information Detection, Extraction and Summarization) program and DUC (Document Understanding Conferences) evaluations.
Since FRUM was developed by DeJong (1982), text summarization has been a very attractive research topic for the Artificial Intelligence community: the construction of programs able to understand and generate language is a traditional concern of AI. This tutorial will give a detailed account of methods and techniques used in automatic text summarization.

About the speaker: Dr. Horacio Saggion is a research fellow in the Natural Language Processing group, Department of Computer Science, University of Sheffield, England, UK. His area of expertise is Text Summarization. He works on information extraction, question answering, and text summarization. He obtained his PhD in 2000 from Université de Montréal, Department d'Informatique et de Recherche Operationnelle; his Master degree from Universidade Estadual de Campinas, Departamento de Computacao in 1995; and his undergraduate degree of "Licenciado" in 1988 from Universidad de Buenos Aires.

Planning under uncertainty with Markov decision processes (8 hours/English and/or Spanish)
L. Enrique Sucar, ITESM Campus Cuernavaca - Mexico
Alberto Reyes, Instituto de Investigaciones Eléctricas - Mexico

Planning in the real world requires to consider the uncertainty in the agents actions and observations. Traditional planning techniques do not take into account this uncertainty, and usually assume a deterministic and fully observable world. They also do not consider the cost or utilities associated to different possible outcomes of the plan. Recently, Markov Decision Processes (MDPs) have developed as a standard for representing uncertainty in decision-theoretic planning. These techniques consider the uncertainty in the world and the cost/utilities of the plans, and can generate a general plan or policy for every possible situation. There are different algorithms for solving a MDP and finding the optimal policy based on value or policy iteration. Although a simple state-based representation  could result in an exponential growth of the state space, recent advances in factored and abstract MDPs help to solve this complexity problem. By using these compact representations, MDPs have been used to solve complex problems in robotics and other domains.
This tutorial will give a general introduction to Markov Decision Processes and illustrate some of their applications.

About the speakers: L. Enrique Sucar is a Professor of Computer Science at the Tecnológico de Monterrey, Campus Cuernavaca, Mexico. Dr. Sucar has a MSc in Electrical Engineering from Stanford University and a PhD degree in Computing from Imperial College, University of London. Dr. Sucar has worked in the area of reasoning under uncertainty for the last 15 years, and in particular in planning and decision making in the last years.
Alberto Reyes is a PhD student in Computer Science at ITESM Cuernavaca. His research is on qualitative Markov decision processes and their application in power plant. He has a MSc in Artificial Intelligence from Universidad Veracruzana-LANIA, 1999, and a BSc in Electrical and Mechanical Engineering, Universidad Veracruzana, 1990.

Organization oriented programming in multi-agent systems (4 hours/English)
Jaime Simão Sichman, University of São Paulo - Brazil

Within a multi-agent system, coarse-grained cognitive agents that can be considered as autonomous can be viewed as entities that present at least 3 different levels of reasoning mechanisms: a domain-specific one, a social one (reasoning about the others) and an organizational one (reasoning about pre-established or emergent duties and responsabilities, linked to social roles and/or social tasks). This tutorial will present some basic notions regarding this latter aspect, called organizational level, covering both theoretical, architectural and implementation levels of abstraction.

About the speaker: Jaime Simão Sichman is an Associate Professor at the Polytechnic School, University of São Paulo. He has a MSc degree in Electrical Engineering from USP Polytechnic School, Brazil, and a PhD degree in Computer Science from INPG, Grenoble, France. His major fields of interest are models of social reasoning in multi-agent systems, cognitive agent models, multi-agent simulation in economic and social sciences, distributed autonomous belief revision and organizational knowledge in mult-agent systems.

Reactive robotics and the Khepera robot (4 hours/Spanish)
Angélica García, Universidad Veracruzana - Mexico
Fernando M. Montes, Universidad Veracruzana - Mexico

At the present, the field of robotics has become popular among several researchers. Industrial robots have a long tradition from automotive applications like painting, arc-welding to delivering industrials components, and collecting fruits. A new branch in robotics has emerged recently, service robots. Some examples of how robots can be used are: floor cleaning, surveillance systems, assistance and carrier tasks in medical domains. These tasks require further refinement of the current state of the art. If we need robots behave in a complete autonomous way, still some prototypes and algorithms can be developed. Commercial tools are used to help the development, of both indoors and outdoors, robotics platforms. The Khepera robot is a clear example of an indoor platform, which can be used with research and educative purposes. On its second version this miniature robot measures about 70 millimeters diameter and 30 millimeters high.

Outline:
    Introduction to reactive robotics
     Review of some commercial robotic platforms, and the available programming languages for these platforms
        Pioneer/Lego MindStorms-/Khepera
    The Khepera robot and its software of control
    Capabilities
        Attachable extension turrets
        Available control software
        Examples of experiments using the Khepera robot
        The KpraSIM simulator
        Foraging
        Social behaviors
        Integration of computer vision
        Final Remarks

About the speakers:  Dr. Fernando M. Montes G. is faculty member at the Artificial Intelligence Department of the University of Veracruz. His research interests are Artificial Intelligence, Artificial Life, Behavior Based Robotics, Action Selection Mechanisms, and Computational Neuroscience. He is Member of the Mexican Association of Researchers (SNI).
MScs Angelica Garcia Vega is senior faculty member at the Artificial Intelligence Department of the University of Veracruz. Her research interests are Behavior-Based Robotics, Artificial Intelligence and Artificial Life.

Advances in word sense disambiguation (4 hours/English)
Ted Pedersen, University of Minnesota - USA
Rada Mihalcea,  University of North Texas - USA

Word sense disambiguation is the problem of identifying the intended meaning (or sense) of a word, based on the context in which it occurs. This is a central problem in natural language processing, and improved approaches have the potential to advance the state of the art in machine traslation, information retrieval, and many other language related problems. This tutorial will introduce the full range of techniques that have been applied to this problem. These include knowledge-intensive methods that take advantage of dictionaries and othe manually created resources, supervised techniques that learn classifiers from training examples, minimally supervised approaches that bootstrap off smal amounts of data, and unsupervised approaches that identify word senses in raw unannotated text.

About the speakers: Ted Pedersen is an Associate Professor of Computer Science at the University of Minnesota, Duluth. He has a  MSc in Computer Science from the University of Arkansas and  PhD degree in Computer Science from the Southern Methodist University in the United States. He has been actively engaged in word sense disambiguation research since 1995.
Rada Mihalcea is an Assistant Professor of Computer Science at the University of North Texas. Her research interests are in lexical semantics, minimally supervised natural language learning, and multilingual natural language processing.

Cross-modal analysis of speech, gestures and gaze through multimedia software tools to capture emotional features in verbal and non verbal communication modalities (4 hours//English)
Anna Esposito, Seconda Università di Napoli - Italy

This course examines  the human communicative behaviour from a psychological and experimental  perspective with the aim to seek for “maybe invariant” emotional features through a cross-modal analysis of verbal and non-verbal modalities. The final goal is to identify procedures and algorithms that define the basis for the implementation of interactive dialog systems.  Interactive dialog systems are computer interfaces represented by lifelike human or animal characters that are capable of performing believable actions and reacting to human users. Such systems may greatly simplify the human-computer interaction.  To allow experimental measurements and psychological analysis I will give a brief introduction to the general acoustic features of emotional speech. Multimedia software tools for the analysis of speech and video sequences (SPEECHSTATION(TM), COOLEDIT(TM), GOLDWAVE(TM), QUICKTIME(c), VIRTUAL DUB, PRAAT, ANVIL, ADOBE PREMIERE) are introduced to facilitate the quantitative and qualitative analysis of the data.

About the speaker: Anna Esposito is Associate Professor in Computer Science in the Department of Psychology at the Second University of Naples. She also teach the course 'Laboratory of Cybernetics' in the Department of Physics at Salerno University since 1997. She is research affiliate in the department of Computer Science and Engineering at Wright State University, Dayton, OH, USA from November 2000; the International Institute for Advanced Scientific Studies (Italy),since 1995, and the Research Laboratory of Electronics, Speech Communication group (MIT) since 1993, where she developed her Ph.D thesis. She obtained the Ph.D degree in Applied Mathematics and Computer Science from Naples University 'Federico II' .