CPSC 609 Foundations of Multi-Agent Systems Winter 2019 Jrg - PowerPoint PPT Presentation

CPSC 609 � Foundations of Multi-Agent Systems � Winter 2019 � Jörg Denzinger � ICT 752 � denzinge@cpsc.ucalgary.ca � http://www.cpsc.ucalgary.ca/~denzinge/ �

1. Introduction � 1.1. Organization � n Lectures: MoWe 15:30 � n Assignments: � ● Midterm paper on a cooperation concept � ● Final paper describing a team of agents that work together rescuing people buried under rubble after a disaster struck a city (agents would be acting within the ARES system) � Multi-Agent Systems Jörg Denzinger

Determining your grade � � n Midterm paper � � � � 30% � n Final paper � � � � 30% � n Oral exam � � � � 40% � Multi-Agent Systems Jörg Denzinger

More Information � n Course webside: � http://pages.cpsc.ucalgary.ca/~denzinge/ � � � � courses/cs609-winter2019.html � n Internet � n Recommended papers � n Talk to me, ask questions, send me email. � Multi-Agent Systems Jörg Denzinger

1.2 Motivation � Robo-Cup Blackboard Avatar Vickrey auction Game theory Intentions Desire � Cooperation Commitment Competition Belief Teamwork ? Contract-net negotiations Market mechanisms Adversary Multi-Agent Systems Jörg Denzinger

MAS and related areas � (in Computer Science) � n Artificial Intelligence � ● Planning � ● Learning � ● Knowledge representation / logic � n Distributed Systems � n Efficient Algorithms � n Security � n Robotic � n (graphical) user interfaces � Multi-Agent Systems Jörg Denzinger

MAS and related areas � (outside Computer Science) � n Business Science / organizational theory / market studies � n Game theory � n Psychology � n Sociology � n Arts � n All kinds of Simulations � n Soccer � n ... � Multi-Agent Systems Jörg Denzinger

Goals of MAS (I) � In general: higher efficiency, faster answers, better solutions � Especially: Synergy � (the whole is more than the sum of its parts) � � Additional goals: � n Adequately modeling given distributions � (functional, spatial, etc.) � n Improving adaptability, extendability, reliability of systems ( F software engineering) � Multi-Agent Systems Jörg Denzinger

Goals of MAS (II) � n Easy integration of existing systems, achieving open systems ( F software engineering) � n More cost efficiency by specialization and distribution of tasks � n More security and safety by autonomy and isolation � n Usage of computer networks for AI applications � n Simulating human approach to hard tasks: � F � computer (program) as colleague or part of a � � colleague � Multi-Agent Systems Jörg Denzinger

Basic problems (I) � n Modeling agents � ● degree of autonomy � ● degree of willingness to collaborate � ● degree of intelligence � ● complexity � n Construction of a community (society, organization) of agents � ● communication � ● control � Multi-Agent Systems Jörg Denzinger

Basic problems (II) � ● different roles � ● division and assignment of tasks � ● distribution/assignment of resources � n Utilizing a community of agents � ● environment � ● constraints/conditions to keep fulfilled � ● efficiency � ● reuse of components � Multi-Agent Systems Jörg Denzinger

Solution approaches (I) � n Adaptation of human forms of organization � ● political structures � ● organization of business companies � ● groups in society � n Negotiation concepts for solving conflicts � ● detection of conflicts � ● determining scope of conflicts � ● protocol for solution process � n Game theoretical concepts, for example utility (payoff) matrixes � Multi-Agent Systems Jörg Denzinger

Solution approaches (II) � n Knowledge representation techniques � F � standards for information exchange, like FIPA � � initiative or KIF � n Different concepts for modeling agents, for example modal logics, state automata, rule based modeling,… � n Tools for realizing MAS � ● MAS languages � ● Test beds, like OLEMAS and ARES � ● Groupware � n ... � Multi-Agent Systems Jörg Denzinger

(Intended) Structure of the course (I) � 1. � Introduction � � 1. � Organization � � 2. � Motivation � � 3. � A little history � 2. � Single-agent systems � � 1. � Formal definitions and properties � � 2. � Modeling of agents � 3. � Multi-agent systems � � 1. � Formal definitions and properties � Multi-Agent Systems Jörg Denzinger

(Intended) Structure of the course (II) � 3. � 2. � Interaction and cooperation concepts � � 3. � Competitive agent environments � 4. � Learning in MAS � � 1. � Learning in general � � 2. � From single-agent to multi-agent learning � � 3. � Examples � 5. Testing MAS � 6. Larger examples (if time permits) � Multi-Agent Systems Jörg Denzinger

1.3. A little history of MAS � 1975 � Hearsay-II, Blackboard systems (Erman, Lesser) � � � Beings: knowledge as interacting experts (Lenat) � 1977 � Actor model, message passing (Hewitt) � 1980 � First workshop on Distributed AI, MIT � � � Contract-net protocol, negotiations as cooperation � � concept (Smith) � 1981 � IEEE Transactions on Systems, Man and � � Cybernetics special issue on DAI: � � � Organizational View on Distributed Systems (Fox) � � � The Scientific Community Metaphor (Kornfeld, � � Hewitt) � Multi-Agent Systems Jörg Denzinger

1981 � Functionally accurate cooperative distributed � � systems (Lesser, Corkill) � 1982 � Synchronization of plans for agents (Rosenschein) � 1983 � Open Systems (Hewitt, de Jong) � 1984 � Modeling of Beliefs, Desires and Intentions � � (Levesque) �� � � Actions in plans for multi-agent systems � � (Georgeff) � 1985 � Pursuit Game as testbed (Benda, Jadannathan, � � Dodhiawala) � 1986 � Cooperation without Communication, Utilizing � � game theory (Rosenschein) � � � Classification dimensions of DAI approaches � � (Sridharan, ed.) � Multi-Agent Systems Jörg Denzinger

1989 � First MAAMAW, Cambridge � � � Reinforcement Learning (Watkins) � 1990 � First MACC � � � Agent-oriented programming (Shoham) � � � Learning in multi-agent systems (Sian) � � � User agents, personal assistant (Hammainen et al.) � � (Pan, Tenenbaum) � 1991 � Knowledge Interchange Format (KIF) (Interlingua � � Working Group) � � � Teamwork concept (Denzinger) � 1993 � Founding Workshop of German DAI group, � � Saarbrücken � 1994 � Softbots (Etzioni, Weld) �� Multi-Agent Systems Jörg Denzinger

1995 � First ICMAS, San Francisco � 1996 � First RoboCup, Kyoto � � � Founding of FIPA-Initiative � 1997 � First Autonomous Agents conference � 1998 � Agents ’ World, Paris � � � Start of the AgentLink program of EU � 1999 � First IAT, Hong Kong � � Internet of Things mentioned by Kevin Ashton � 2001 � First RoboCupRescue competition � 2002 � First AAMAS in Italy � � � First Trading Agent Competition (TAC) � First Workshop on Teaching MAS at AAMAS � Multi-Agent Systems Jörg Denzinger

2006 First ART Competition (Using the Agent � � Reputation and Trust Testbed) � � Most influential paper award at AAMAS � First COMMA (Computational Models of � Argument) Conference in Liverpool � 2007 � First SASO (Self-Adaptive and Self-Organizing � � Systems) Conference at MIT � 2016 � IAT rolled into Web Intelligence conference � � ICA started as alternative � In general, trend towards specialized conferences and sub-areas “ breaking away ” from the field � Other disciplines making use of MAS: security, Internet of Things, etc. � � Multi-Agent Systems Jörg Denzinger