## Coalitions Among Computationally Bounded Agents (1997)

### Cached

### Download Links

- [www.lirmm.fr]
- [jmvidal.cse.sc.edu]
- [mas.cs.umass.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | Artificial Intelligence |

Citations: | 182 - 25 self |

### BibTeX

@ARTICLE{Sandholm97coalitionsamong,

author = {Tuomas W. Sandholm and Victor R. Lesser},

title = {Coalitions Among Computationally Bounded Agents},

journal = {Artificial Intelligence},

year = {1997},

volume = {94},

pages = {99--137}

}

### Years of Citing Articles

### OpenURL

### Abstract

This paper analyzes coalitions among self-interested agents that need to solve combinatorial optimization problems to operate e ciently in the world. By colluding (coordinating their actions by solving a joint optimization prob-lem) the agents can sometimes save costs compared to operating individually. A model of bounded rationality is adopted where computation resources are costly. It is not worthwhile solving the problems optimally: solution quality is decision-theoretically traded o against computation cost. A normative, application- and protocol-independent theory of coalitions among bounded-rational agents is devised. The optimal coalition structure and its stability are signi cantly a ected by the agents ' algorithms ' performance pro les and the cost of computation. This relationship is rst analyzed theoretically. Then a domain classi cation including rational and bounded-rational agents is in-troduced. Experimental results are presented in vehicle routing with real data from ve dispatch centers. This problem is NP-complete and the instances are so large that|with current technology|any agent's rationality is bounded by computational complexity. 1

### Citations

2991 |
Game Theory
- Fudenberg, Tirole
- 1991
(Show Context)
Citation Context ...h bidder is free to submit one bid to take responsibility for a task, which isawarded to the lowest price bidder at the price of his bid. The analog of a protocol is called a mechanism in game theory =-=[11, 24]-=-. 2sguarantees that each agent's desired local strategy is best for that agent|and thus the agent will use it. The normative approach is required in designing robust nonmanipulable multiagent systems ... |

1668 |
Microeconomic Theory
- Mas-Colell, Whinston, et al.
- 2004
(Show Context)
Citation Context ... may be constructed by separate designers and/or may represent di erent real world parties. Interactions of self-motivated agents have been widely studied in microeconomics| especially in game theory =-=[28, 11, 24, 33]-=-. Most of that work assumes perfect rationality of the agents [49, 18], e.g., awless and costless deduction. We extend the normative approach of game theory to settings where the agents lack full rati... |

740 |
Equilibrium Points in n-person Games
- Nash
- 1950
(Show Context)
Citation Context ... postulates desirable stability properties of the outcomes, but does not guarantee stability of the strategies that lead to those outcomes. 16 One alternative solution concept is the Nash equilibrium =-=[29, 24]-=-. It guarantees stability in the sense that no agent alone is motivated to deviate by changing its strategy given that others do not deviate. Often the Nash equilibrium is too weak because subgroups o... |

479 |
The Art and Science of Negotiation
- Raiffa
- 1982
(Show Context)
Citation Context ...presents related research, and Section 8 concludes and describes future research directions. 2 Computation unit cost and algorithm as limits to rationality Coalition formation has been widely studied =-=[20, 56, 32, 53, 52, 59, 22]-=-, but to our knowledge, only among rational agents which can solve the coalition's optimization problem exactly, immediately, and without computation cost. This section describes how our model differs... |

371 |
A Course in Microeconomic Theory
- Kreps
- 1990
(Show Context)
Citation Context ...h bidder is free to submit one bid to take responsibility for a task, which isawarded to the lowest price bidder at the price of his bid. The analog of a protocol is called a mechanism in game theory =-=[11, 24]-=-. 2sguarantees that each agent's desired local strategy is best for that agent|and thus the agent will use it. The normative approach is required in designing robust nonmanipulable multiagent systems ... |

299 |
Zlotkin G. “Rules of Encounter
- Rosenschein
- 1994
(Show Context)
Citation Context ... approach is usually descriptive: the main question is what social outcomes follow given the protocol and assuming that the agents use the imposed strategies. On the other hand, in multiagent systems =-=[34, 23, 5, 47, 41, 44]-=-, the agents are provided with an interaction protocol, but each agent will choose its own strategy. A self-interested agent will choose the best strategy for itself, which cannot be explicitly impose... |

290 | A market-oriented programming environment and its application to distributed multicommodity flow problems
- Wellman
- 1993
(Show Context)
Citation Context ...ments are made regarding each contract before new contracts take place. An agent updates its approximate local solution after each task transfer. In general equilibrium market mechanisms suchasWalras =-=[57]-=-, non-manipulative agents iterate over the allocation of resources and tasks, and payments are usually made only after a nal solution has been reached. Unlike our work, general equilibrium methods are... |

281 |
An implementation of contract net protocol based on marginal cost calculations
- Sandholm
- 1993
(Show Context)
Citation Context ...ey represent one week delivery order and vehicle data. 5 The collected data is characterized in Table 1. Our prior work has already focused on di erent aspects of automated negotiation in this domain =-=[37, 40, 44, 38, 39, 26, 45, 42]-=-, and lately other researchers have studied an almost identical problem, yet with randomly generated instances and with a non-normative approach [9]. Also, simpler routing problems have often been use... |

258 |
Acceptable points in general cooperative n-person games
- Aumann
- 1959
(Show Context)
Citation Context ... deviate by changing its strategy given that others do not deviate. Often the Nash equilibrium is too weak because subgroups of agents can deviate in a coordinated manner. The Strong Nash equilibrium =-=[1]-=- is a solution concept for NFGs that guarantees more stability. It requires that there is no subgroup that can deviate by changing their strategies jointly in a manner that increases the payo of all o... |

227 | Issues in automated negotiation and electronic commerce: Extending the contract net framework
- Sandholm, Lesser
- 1995
(Show Context)
Citation Context ...ore, the coalition structure is stable against all o ers. Finally, his 2-agent auction is manipulable and computationally ine cient. Ketchpel's method is related to a contracting protocol of Sandholm =-=[40, 46]-=- (Traconet) where agents construct the global solution by contracting a small number of tasks at a time, and payments are made regarding each contract before new contracts take place. An agent updates... |

198 |
Spieltheoretische Behandlung eines Oligopolmodells mit Nachfrageträgheit,” Zeitschrift für die gesamte Staatswissenschaft
- Selten
- 1965
(Show Context)
Citation Context ...tability for the beginning of the game. To ensure stability throughout the game, the equilibria should prevail in subtrees of the game tree also. To guarantee this, sequential [24] or subgame perfect =-=[50, 24]-=- re nements of the above solution concepts should be used. The strategic solution concepts presented above provide a rigorous tool for analyzing general games (NFGs), and they also extend the strategi... |

191 | Reasoning about beliefs and actions under computational resource constraints
- Horvitz
- 1987
(Show Context)
Citation Context ...ecided how much CPU time rS it will allocate to a computation, it can design an algorithm that will nd a solution of cost cS(rS). The design-to-time framework is used instead of the anytime framework =-=[43, 4, 17, 58]-=- because to devise a normative theory of self-interested agents, the possibility that they design their algorithms to time has to be accounted for. With deterministic performance pro les, for any desi... |

189 |
Do the right thing
- Russell, Wefald
- 1991
(Show Context)
Citation Context ...eliberation controller composes the optimal sequence of base level computation actions since even the most advanced methods for such composition rely on assumptions that often do not hold in practice =-=[36]-=-. Assuming that the metalevel exactly and costlessly predicts the solution cost is more realistic than assuming optimality of the base level, but it still does not match reality exactly. In practice t... |

179 | Frontiers of Electronic Commerce - Kalakota, Whinston - 1996 |

168 | Cooperative Distributed Problem Solving
- Durfee, Lesser, et al.
- 1989
(Show Context)
Citation Context ...dependent businesses or individuals. In building computer support for coalition formation in such settings, the issue of self-interest has to be dealt with. In cooperative distributed problem solving =-=[7, 5]-=-, the system designer imposes an interaction protocol 2 and a strategy (a mapping from state history to action; a way to use the protocol) for each agent. The approach is usually descriptive: the main... |

164 | Deliberation scheduling for problem solving in time-constrained environments
- Boddy, Dean
- 1994
(Show Context)
Citation Context ...ecided how much CPU time rS it will allocate to a computation, it can design an algorithm that will nd a solution of cost cS(rS). The design-to-time framework is used instead of the anytime framework =-=[43, 4, 17, 58]-=- because to devise a normative theory of self-interested agents, the possibility that they design their algorithms to time has to be accounted for. With deterministic performance pro les, for any desi... |

152 |
Negotiation under time constraints
- Kraus, Wilkenfeld, et al.
- 1995
(Show Context)
Citation Context ... approach is usually descriptive: the main question is what social outcomes follow given the protocol and assuming that the agents use the imposed strategies. On the other hand, in multiagent systems =-=[34, 23, 5, 47, 41, 44]-=-, the agents are provided with an interaction protocol, but each agent will choose its own strategy. A self-interested agent will choose the best strategy for itself, which cannot be explicitly impose... |

139 |
On Balanced Sets and Cores
- Shapley
- 1967
(Show Context)
Citation Context ...are not BRSUB, the BRC is sometimes empty. The condition C 6= ; can be converted into necessary and su cient conditions on the v R S values in games where the grand coalition maximizes social welfare =-=[51, 6]-=-. We convert the condition BRC(ccomp) 6= ; into conditions on the vS(ccomp) values analogously. Let B1; :::; Bp be distinct, nonempty, proper subsets of A. The set B = fB1; :::; Bpg is called balanced... |

124 | Limitations of the Vickrey auction in computational multiagent systems
- Sandholm
- 1996
(Show Context)
Citation Context ... approach is usually descriptive: the main question is what social outcomes follow given the protocol and assuming that the agents use the imposed strategies. On the other hand, in multiagent systems =-=[34, 23, 5, 47, 41, 44]-=-, the agents are provided with an interaction protocol, but each agent will choose its own strategy. A self-interested agent will choose the best strategy for itself, which cannot be explicitly impose... |

118 | Optimal composition of real-time systems
- Zilberstein, Russell
- 1996
(Show Context)
Citation Context ...itial solutions are used (their computation requirements are assumed negligible). 2.1 Discussion of this model of bounded rationality Conceptually we allow the agents to use design-to-time algorithms =-=[12, 58, 13]-=-: once an agent has decided how much CPU time rS it will allocate to a computation, it can design an algorithm that will nd a solution of cost cS(rS). The design-to-time framework is used instead of t... |

117 | Design-to-time real-time scheduling
- Garvey, Lesser
- 1993
(Show Context)
Citation Context ...itial solutions are used (their computation requirements are assumed negligible). 2.1 Discussion of this model of bounded rationality Conceptually we allow the agents to use design-to-time algorithms =-=[12, 58, 13]-=-: once an agent has decided how much CPU time rS it will allocate to a computation, it can design an algorithm that will nd a solution of cost cS(rS). The design-to-time framework is used instead of t... |

116 |
Coalition-proof Nash equilibria I: Concepts
- Bernheim, Peleg, et al.
- 1987
(Show Context)
Citation Context ...s do not deviate from the original solution. The Strong Nash equilibrium is often too strong a solution concept: in many games no such equilibria exist. Recently, the Coalition-Proof Nash equilibrium =-=[2, 3]-=- for NFGs has been suggested as a partial remedy to the nonexistence problem of the Strong Nash equilibrium. This solution concept requires that there is no subgroup that can make a bene cial deviatio... |

109 | Negotiation among self-interested computationally limited agents
- Sandholm
- 1996
(Show Context)
Citation Context ...ey represent one week delivery order and vehicle data. 5 The collected data is characterized in Table 1. Our prior work has already focused on di erent aspects of automated negotiation in this domain =-=[37, 40, 44, 38, 39, 26, 45, 42]-=-, and lately other researchers have studied an almost identical problem, yet with randomly generated instances and with a non-normative approach [9]. Also, simpler routing problems have often been use... |

105 | Advantages of a leveled commitment contracting protocol: Extended version. Computer Science
- Sandholm, Lesser
- 1995
(Show Context)
Citation Context |

95 | Task allocation via coalition formation among autonomous agents
- Shehory, Kraus
(Show Context)
Citation Context ...problem, yet with randomly generated instances and with a non-normative approach [9]. Also, simpler routing problems have often been used as example applications in recent multiagent systems research =-=[34, 59, 52, 57]-=-. Table 1: One week of real vehicle and delivery data used in the experiments. Dispatch Number of Number of Average center delivery orders vehicles delivery length 1 65 10 121 km 2 200 13 169 km 3 82 ... |

85 | Provably BoundedOptimal Agents - Russell, Subramanian - 1995 |

77 | Coalition formation among bounded rational agents
- Sandholm, Lesser
- 1996
(Show Context)
Citation Context ...ein Foundation, Jenny and Antti Wihuri Foundation, Honkanen Foundation, Ella and George Ehrnrooth Foundation, and the Thanks to Scandinavia Foundation. A short early version of this paper appeared in =-=[44]-=-. 1s1 Introduction Automated negotiation systems with self-interested agents are becoming increasingly important. One reason for this is the technology push of a growing standardized communication inf... |

60 | Forming coalitions in the face of uncertain rewards
- Ketchpel
- 1994
(Show Context)
Citation Context ...presents related research, and Section 8 concludes and describes future research directions. 2 Computation unit cost and algorithm as limits to rationality Coalition formation has been widely studied =-=[20, 56, 32, 53, 52, 59, 22]-=-, but to our knowledge, only among rational agents which can solve the coalition's optimization problem exactly, immediately, and without computation cost. This section describes how our model di ers ... |

57 |
Theories of Coalition Formation. Lawrence Erlbaum Associates
- Kahan, Rapoport
- 1984
(Show Context)
Citation Context ...presents related research, and Section 8 concludes and describes future research directions. 2 Computation unit cost and algorithm as limits to rationality Coalition formation has been widely studied =-=[20, 56, 32, 53, 52, 59, 22]-=-, but to our knowledge, only among rational agents which can solve the coalition's optimization problem exactly, immediately, and without computation cost. This section describes how our model differs... |

56 | A Model For Cooperative Transportation Scheduling
- Fischer, Müller, et al.
- 1995
(Show Context)
Citation Context ...tiation in this domain [37, 40, 44, 38, 39, 26, 45, 42], and lately other researchers have studied an almost identical problem, yet with randomly generated instances and with a non-normative approach =-=[9]-=-. Also, simpler routing problems have often been used as example applications in recent multiagent systems research [34, 59, 52, 57]. Table 1: One week of real vehicle and delivery data used in the ex... |

56 |
On complexity as bounded rationality
- Papadimitriou, Yannakakis
- 1994
(Show Context)
Citation Context ...e examined the e ects of computational limits on rational play in settings where agents play a combinatorially trivial game, but complexity stems from numerous repetitions of that same game, see e.g. =-=[31]-=-. 3sEach vehicle has a maximum load weight constraint. These di er among vehicles. Each vehicle has a maximum load volume constraint. These also di er among vehicles. Each vehicle has the same maximum... |

51 |
Models of bounded rationality, volume 2
- Simon
- 1982
(Show Context)
Citation Context ...osts. However, if the problem is hard and the instance is large, it is unrealistic to assume that it can be solved without deliberation costs. This paper adopts a speci c model of bounded rationality =-=[54, 15]-=-, where each agent has to pay for the computational resources (CPU cycles) that it uses for deliberation. A xed computation cost ccomp 0 per CPU time unit is assumed. 7 The domain cost associated with... |

46 | Equilibrium analysis of the possibilities of unenforced exchange in multiagent systems
- Sandholm, Lesser
- 1995
(Show Context)
Citation Context ...ey represent one week delivery order and vehicle data. 5 The collected data is characterized in Table 1. Our prior work has already focused on di erent aspects of automated negotiation in this domain =-=[37, 40, 44, 38, 39, 26, 45, 42]-=-, and lately other researchers have studied an almost identical problem, yet with randomly generated instances and with a non-normative approach [9]. Also, simpler routing problems have often been use... |

44 | A Survey of Research in Deliberative Real-Time Artificial Intelligence. UMass Computer Science
- Garvey, Lesser
- 1983
(Show Context)
Citation Context ...itial solutions are used (their computation requirements are assumed negligible). 2.1 Discussion of this model of bounded rationality Conceptually we allow the agents to use design-to-time algorithms =-=[12, 58, 13]-=-: once an agent has decided how much CPU time rS it will allocate to a computation, it can design an algorithm that will nd a solution of cost cS(rS). The design-to-time framework is used instead of t... |

40 |
Twenty-seven principles of rationality
- Good
- 1971
(Show Context)
Citation Context ...sts. However, if the problem is hard and the instance is large, it is unrealistic to assume that it can be solved without deliberation costs. This paper adopts a specific model of bounded rationality =-=[54, 15]-=-, where each agent has to pay for the computational resources (CPU cycles) that it uses for deliberation. A fixed computation cost c comps0 per CPU time unit is assumed. 7 The domain cost associated w... |

37 |
A language and protocol to support intelligent agent interoperability
- Finin, Fritzson, et al.
- 1992
(Show Context)
Citation Context ...ystems with self-interested agents are becoming increasingly important. One reason for this is the technology push of a growing standardized communication infrastructure|Internet, WWW, NII, EDI, KQML =-=[8]-=-, FIPA, Telescript [14], Java, etc|over which separately designed agents belonging to di erent organizations can interact in an open environment in real-time and safely carry out transactions. The sec... |

37 | Coalition formation processes with belief revision among bounded-rational self-interested agents
- Tohmé, Sandholm
- 1999
(Show Context)
Citation Context ...els of bounded rationality besides these two also deserve attention. Our current work includes analyzing the interplay of dynamic coalition formation and belief revision among bounded-rational agents =-=[55]-=-. Extensions of our work include generalizing the methods of this paper to agents with di erent and probabilistic performance pro les, as well as anytime algorithms where the performance pro les are c... |

35 | Monitoring the Progress of Anytime Problem-Solving
- Hansen, Zilberstein
- 1996
(Show Context)
Citation Context ... general, for optimal meta-reasoning, the remaining part of a probabilistic performance pro le should be conditioned on the algorithm's performance on that problem instance on previous CPU time steps =-=[43, 58,16]-=-. 10sis divided among the agents in the coalition as will be presented later. To summarize, with our model of bounded rationality it is best to centralize computation within each coalition but computa... |

30 | Self-fulfilling bias in multiagent learning
- Hu, Wellman
- 1995
(Show Context)
Citation Context ...uilibrium where the agent's maximal gain from speculation materializes [48]. Speculative behavior in general equilibrium markets has recently been studied in the context of learning by Hu and Wellman =-=[19]-=-. Shehory and Kraus [53] analyze coalition formation among rational agents with perfect information in CFGs that are not necessarily superadditive. Their protocol guarantees that if agents follow it (... |

27 | On the Gains and Losses of Speculation in Equilibrium Markets
- Sandholm, Ygge
- 1997
(Show Context)
Citation Context ...et prices. Recently, Sandholm and Ygge have devised general insincere strategies that allow an agent to drive the market to an equilibrium where the agent's maximal gain from speculation materializes =-=[48]-=-. Speculative behavior in general equilibrium markets has recently been studied in the context of learning by Hu and Wellman [19]. Shehory and Kraus [53] analyze coalition formation among rational age... |

26 |
Paradoxes of Rationality: Theory of Metagames and Political Behavior
- Howard
- 1971
(Show Context)
Citation Context ...orld parties. Interactions of self-motivated agents have been widely studied in microeconomics| especially in game theory [28, 11, 24, 33]. Most of that work assumes perfect rationality of the agents =-=[49, 18]-=-, e.g., awless and costless deduction. We extend the normative approach of game theory to settings where the agents lack full rationality because they cannot enumerate or evaluate all alternative solu... |

17 |
The Limits of Rationality
- Cook, Levi, et al.
- 1990
(Show Context)
Citation Context ...orld parties. Interactions of self-motivated agents have been widely studied in microeconomics| especially in game theory [28, 11, 24, 33]. Most of that work assumes perfect rationality of the agents =-=[49, 18]-=-, e.g., awless and costless deduction. We extend the normative approach of game theory to settings where the agents lack full rationality because they cannot enumerate or evaluate all alternative solu... |

16 |
A strategy for decreasing the total transportation costs among area-distributed transportation centers
- Sandholm
- 1991
(Show Context)
Citation Context |

16 |
Stability: Mechanisms for Coalition Formation in Task Oriented Domains
- Zlotkin, Rosenschein, et al.
- 1994
(Show Context)
Citation Context ...problem, yet with randomly generated instances and with a non-normative approach [9]. Also, simpler routing problems have often been used as example applications in recent multiagent systems research =-=[34, 59, 52, 57]-=-. Table 1: One week of real vehicle and delivery data used in the experiments. Dispatch Number of Number of Average center delivery orders vehicles delivery length 1 65 10 121 km 2 200 13 169 km 3 82 ... |

15 |
Telescript technology: The foundation for the electronic marketplace
- Magic, Inc
- 1994
(Show Context)
Citation Context ...ested agents are becoming increasingly important. One reason for this is the technology push of a growing standardized communication infrastructure|Internet, WWW, NII, EDI, KQML [8], FIPA, Telescript =-=[14]-=-, Java, etc|over which separately designed agents belonging to di erent organizations can interact in an open environment in real-time and safely carry out transactions. The second reason is strong ap... |

13 |
Coalition-proof Nash equilibria II: Applications
- Bernheim, Whinston
- 1987
(Show Context)
Citation Context ...s do not deviate from the original solution. The Strong Nash equilibrium is often too strong a solution concept: in many games no such equilibria exist. Recently, the Coalition-Proof Nash equilibrium =-=[2, 3]-=- for NFGs has been suggested as a partial remedy to the nonexistence problem of the Strong Nash equilibrium. This solution concept requires that there is no subgroup that can make a bene cial deviatio... |

13 |
An effective heuristic procedure for the traveling salesman problem
- Lin, Kemighan
- 1971
(Show Context)
Citation Context ... completing an iterative re nement algorithm by running an exhaustive complete algorithm after the re nement phase. Another example is switching from using one re nement operator (e.g., 2-swap in TSP =-=[25, 40]-=-) to using another re nement operator (e.g., 3-swap in TSP). Furthermore, re nements often decrease solution cost in a step-wise, noncontinuous manner rendering the performance pro les locally nonconv... |

11 | Utility-based termination of anytime algorithms
- Sandholm, Lesser
- 1994
(Show Context)
Citation Context ...ecided how much CPU time rS it will allocate to a computation, it can design an algorithm that will nd a solution of cost cS(rS). The design-to-time framework is used instead of the anytime framework =-=[43, 4, 17, 58]-=- because to devise a normative theory of self-interested agents, the possibility that they design their algorithms to time has to be accounted for. With deterministic performance pro les, for any desi... |

11 | RobertWilson, "Sequential Equilibrium," Econometrica 50:86394 - Kreps - 1982 |

10 |
Theories of Coalition Formation,” Lawrence Erlbaum Associates
- Rapoport
- 1984
(Show Context)
Citation Context ...presents related research, and Section 8 concludes and describes future research directions. 2 Computation unit cost and algorithm as limits to rationality Coalition formation has been widely studied =-=[20, 56, 32, 53, 52, 59, 22]-=-, but to our knowledge, only among rational agents which can solve the coalition's optimization problem exactly, immediately, and without computation cost. This section describes how our model di ers ... |