Projects

Research Methodologies Tutorial

Ever wondered what agents and multiagent systems were about? How the multitude or sub-areas represented in the community fit together? Why there are so many research styles and what distinguishes them? How to enter the field as an outsider or how to maximise the impact of your own research within the field?' If these and similar questions interest you and you are attending AAMAS 2011, sign up for the...

Half-Day Tutorial on Multiagent Systems Research Methodologies at AAMAS 2011, 2nd May 2011, Taipei, Taiwan

Overview

This full-(half-)day tutorial provides an introduction to the field of agents and multiagent systems and an in-depth discussion of its methodological foundations. Starting from an overview of the history and state of the art in the field, we will review main research methods, approaches to evaluating agents research, and provide guidance for planning, structuring, and conducting high- quality research pro jects so as to avoid methodological pitfalls and maximise impact. Moreover, the tutorial will provide space and time for method reflexion and debate on different approaches to agents research, and for “taking stock” of the state of the field.

Content

  1. What is agents research?
    • A brief history of agents: AI and distributed AI, practical reasoning & logic-based agents, decision-theoretic AI and optimal agents, the birth of game-theoretic AI, agents and the Web, recent developments
    • The struggle for definitions: problems with concepts of agency and its connotations, the breadth of the field, what is really new about agents
    • Demarcation lines: relationships to other disciplines (generally: computer science, AI, economics, social sciences, cognitive science; specifically: knowledge representation and reasoning, planning, machine learning, robotics, service-oriented computing, software engineering, complex systems, “general” AI)
    • Agent research topics: agent architectures, agent communication, reaching agreement, working together, coordination methods, distributed rational decision making, virtual agents, social simulation, agent-oriented software engineering, multi-robot systems, multiagent learning
  2. Agents research methods
    • Formalising agent systems: abstract mathematical models, MDPs and game-theoretic models vs. relational and logic-based representations
    • Single-agent reasoning models: decision-theoretic vs. practical reasoning models
    • Multiagent reasoning models: game-theoretic vs. cooperative models
    • Multiagent approaches to standard AI problems: search, inference, planning, learning
    • Solutions for novel problems: negotiation, coordination, trust, mechanism design
    • Agent systems engineering: agent programming languages, agent communication mechanisms, methodologies, standards and tools
    • Human-oriented agent systems: imitating, dealing with, and simulating humans
  3. Evaluating agents research
    • Theories and theorems: formal modelling, analytical proofs, “theories without theorems”, conceptual architectures and frameworks, “computational philosophy”
    • Simulation and testing: performance metrics, benchmarks and the lack thereof, parameter chaos, happy graphs
    • Human experimentation: controlled experiments, pilot studies, getting data
    • Systems building: the “instance trap”, case studies and coverage, complex theories vs. simplistic systems, solving a “real” real-world problem
    • Interdisciplinary research: operating at the frontiers of different disciplines
  4. How to do research in agents
    • Good and bad hypotheses: applying the scientific method
    • Reinventing the wheel: incremental vs. exploratory research
    • Going back and forth: method-driven vs. goal-driven research
    • In search of rigour: how “dirty” should research be?
    • The enemy never sleeps: why non-agent research should matter to you
    • Look Mom, no hands: the problem of evaluation
    • In the interest of humanity: applications and impact

Target Audience

The primary target audience of this tutorial are doctoral students, researchers new to the field, and practitioners who are interested in using agents methods in their applications. For more experienced researchers in the field, it can provide an interesting forum to reflect on the state of the field, its foundations, and to get a concise overview of current trends and developments. No specific technical background is required.

About the Presenter

Michael Rovatsos is a Lecturer at the School of Informatics of the University of Edinburgh, where he leads the Agents Group within the Centre for Intelligent Systems and their Applications. He holds a PhD in Informatics from the Technical University of Munich (2004), and a Diploma in Informatics from the University of Saarbruecken (1999). His research is in agents and multiagent systems, and he has (co-)authored over 50 papers in this area. He has in-depth knowledge of the international agents research landscape: He has published and presented papers at all but one of the previous AAMAS/ICMAS conferences since 2000. He has taught University-level courses on agent-related topics for five years, visited many international research groups in the area, and has been actively involved in the organisation of over 35 international agent-related events.

Agents Group

Wiki

edit SideBar

Blix theme adapted by David Gilbert, powered by PmWiki