Citation Context

... global utility it is necessary to avoid having the agents work at cross-purposes lest phenomena like the Tragedy of the Commons (TOC) occur, in which individual avarice works to lower global utility =-=[12]-=-. One way to avoid such phenomena is by modifying the agents' utility functions via punitive legislation. A real world example of an attempt to make such a modification was the cration of anti-trust r...

Citation Context

...tistical mechanics; ffl computational ecologies; ffl game theory , in particular, evolutionary game theory. Previous MAS's most similar to a COIN include those where agents use reinforcement learning =-=[8, 13]-=-, and/or where agents actively attempt to model the behavior of other agents [14]. 1 In this paper we introduce some of the concepts from the COIN framework, and then present experiments testing those...

Citation Context

...tistical mechanics; ffl computational ecologies; ffl game theory , in particular, evolutionary game theory. Previous MAS's most similar to a COIN include those where agents use reinforcement learning =-=[8, 13]-=-, and/or where agents actively attempt to model the behavior of other agents [14]. 1 In this paper we introduce some of the concepts from the COIN framework, and then present experiments testing those...

Citation Context

...convergence using the WL reward to be far quicker than that using the GR reward. The GR reward does eventually converge to the global optimum. This is in agreement with the results obtained by Crites =-=[9]-=- for the bank of elevators control problem. However, when ff = [0 0 0 7 0 0 0] the GR reward converged in 1250 weeks. This is more than 4 times the convergence time for the WL reward. When ff = [1 1 1...

Citation Context

...ize each agents' local reward function) to ensure that the agents do not work at cross purposes. The problem we chose for this purpose is a more challenging variant of Arthur's bar attendance problem =-=[1, 7, 17]-=-. In this problem, agents have to determine which night in the week to attend a bar. The problem is set up so that if either too few people attend (boring evening) or too many people attend (crowded e...

Citation Context

... the entire system; global performance is "robust"; one can scale up to very large systems; and one can maximally exploit the power of machine learning. We use the term COllective INtelligen=-=ce (COIN) [21, 22, 23]-=- to refer to either MAS's designed in this way, or (in the case of naturally occurring MAS's) to MAS's investigated from this perspective. The COIN framework is related to many other fields. (See [21]...

Citation Context

...olutionary game theory. Previous MAS's most similar to a COIN include those where agents use reinforcement learning [8, 13], and/or where agents actively attempt to model the behavior of other agents =-=[14]-=-. 1 In this paper we introduce some of the concepts from the COIN framework, and then present experiments testing those concepts. The restricted version of the framework presented here is not sufficie...

Citation Context

... the entire system; global performance is "robust"; one can scale up to very large systems; and one can maximally exploit the power of machine learning. We use the term COllective INtelligen=-=ce (COIN) [21, 22, 23]-=- to refer to either MAS's designed in this way, or (in the case of naturally occurring MAS's) to MAS's investigated from this perspective. The COIN framework is related to many other fields. (See [21]...

Citation Context

...estigate the real-world applicability of the core concepts of that framework via two computer experiments: we show that our COINs perform near optimally in a difficult variant of Arthur's bar problem =-=[1]-=- (and in particular avoid the tragedy of the commons for that problem), and we also illustrate optimal performance for our COINs in the leader-follower problem. 1 INTRODUCTION In this paper we are int...

Citation Context

...ize each agents' local reward function) to ensure that the agents do not work at cross purposes. The problem we chose for this purpose is a more challenging variant of Arthur's bar attendance problem =-=[1, 7, 17]-=-. In this problem, agents have to determine which night in the week to attend a bar. The problem is set up so that if either too few people attend (boring evening) or too many people attend (crowded e...

Citation Context

... the entire system; global performance is "robust"; one can scale up to very large systems; and one can maximally exploit the power of machine learning. We use the term COllective INtelligen=-=ce (COIN) [21, 22, 23]-=- to refer to either MAS's designed in this way, or (in the case of naturally occurring MAS's) to MAS's investigated from this perspective. The COIN framework is related to many other fields. (See [21]...