ijcai 2011 presentation
TRANSCRIPT
COLLUSION RESISTANT REPUTATION
MECHANISM FOR MULTI AGENT
SYSTEMS
Babak Khosravifar, Jamal Bentahar, Maziar Gomrokchi
and Mahsa Alishahi
Concordia University, Montreal, Canada
1
OUTLINE
Preliminaries
The Model
Results
Conclusion
References
2
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
OUTLINE
Preliminaries
The Model
Results
Conclusion
References
3
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
PRELIMINARIES
Agent
Environment
Agent
see action
next state
4
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
PRELIMINARIES
Agent
Multi agent system
5
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
PRELIMINARIES
Agent
Multi agent system
Knowledge
Trust and Reputation
Web service agent
Consumer agent
Collusion
6
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
Manager
Task Announcement
NODE ISSUES TASK ANNOUNCEMENT
7
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
Manager
Manager
Manager
PotentialContractor
IDLE NODE LISTENING TO TASK
ANNOUNCEMENTS
8
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
Manager
PotentialContractor
Bid
NODE SUBMITTING A BID
9
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
Manager
PotentialContractor
PotentialContractor
Bids
MANAGER LISTENING TO BIDS
10
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
Manager
Contractor
Award
MANAGER MAKING AN AWARD
11
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
Manager
Contractor
Contract
CONTRACT ESTABLISHED
12
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
OUTLINE
Preliminaries
The Model
Results
Conclusion
References
13
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
THE MODEL
Consumer/Provider strategy profile
Collusion Benefits
Consumer agent ( ε )
Web service agent ( )
Controller agent’s investigation parameters
Analyzing feedback window ( )
Detecting fake feedback ( )
Penalty ( )
W
cw
cdf
Pn
14
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
THE MODEL
Four possible scenarios
Actual collusion is detected
Actual collusion is ignored
Truthful action is penalized
Truthful action is detected
15
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
OUTLINE
Preliminaries
The Model
Results
Conclusion
References
16
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
RESULTS
In repeated game with decision making process, if
the falsely detected feedback is more that correctly
detected ones, web service and consumer agents
choose collusion as dominant strategy.
Penalizing the collusion is Pure Strategy Nash
Equilibrium.
17
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
RESULTS
Penalizing probability
Expected Payoffs
18
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
RESULTS
Estimated penalizing probability
In mixed strategy repeated games, there is a
threshold μ such that if qw > μ acting truthful would be
the dominant strategy.
19
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
RESULTS
If the estimated probability of penalizing exceeds
the obtained threshold, acting truthful and not being
penalized would be the Mixed Strategy Nash
Equilibrium.
A collusion resistant reputation mechanism is
achieved when the controller agent maximizes the
following value.
20
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
RESULTS
21
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
OUTLINE
Preliminaries
The Model
Results
Conclusion
References
22
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
CONCLUSION
Reputation mechanism
Collusion analysis
Collusion resistant structure
Best response analysis
Three player game
Learning methods
MDP/PO-MDP
23
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi
REFERENCES
Archie Chapman, Alex Rogers, Nicholas Jennings, and David Leslie. A unifying framework for iterative approximate best response algorithms for distributed constraint optimization problems. Knowledge Engineering Review (in press), 2011.
Radu Jurca and Boi Faltings. Collusion-resistant, incentive-compatible feedback payments. In Proc. of the ACM Conf. on E-Commerce, pages 200–209, 2007.
Radu Jurca, Boi Faltings, andWalter Binder. Reliable QoS monitoring based on client feedback. In Proc. of the 16’th Int. World Wide Web Conf., pages 1003–1011, 2007.
Georgia Kastidou, Kate Larson, and Robin Cohen. Exchanging reputation information between communities: A payment-function approach. In Proc. of the 21st Int. Joint Conf. on Artificial Intelligence (IJCAI), pages 195–200, 2009.
Babak Khosravifar, Jamal Bentahar, Philippe Thiran, Ahmad Moazin, and Addrien Guiot. An approach to incentive-based reputation for communities of web services. In Proc. of IEEE 7’th Int. Con. on Web Services (ICWS), pages 303–310, 2009.
Babak Khosravifar, Jamal Bentahar, Ahmed Moazin, and Philippe Thiran. On the reputation of agent-based web services. In Proc. of the 24’th Conf. on Artificial Intelligence (AAAI), pages 1352–1357, 2010.
E. Michael Maximilien and Munindar P. Singh. Conceptual model of web service reputation. SIGMOD Record, ACM Special Interest Group on Management of Data, 31(4):36– 41, 2002.
George Vogiatzis, Ian MacGillivray, and Maria Chli. A probabilistic model for trust and reputation. In Proc. of 9’th Int. Conf. on Autonomous Agent and Multi Agent Systems (AAMAS), pages 225–232, 2010.
24
Collusion Resistant Reputation Mechanism for Multi Agent Systems B. Khosravifar, J. Bentahar, M. Gomrokchi, M. Alishahi