Y. Shoham and M. Tennenholtz, (1994). Co-learning and the evolution of social activity. Technical Report STAN-CS-TR-94-1511, Stanford University.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... and conventions break the customary ones How should one reconcile stability and innovation Genetic algorithms bridge this hiatus by introducing mutation (for a discussion on different mechanisms for obtaining innovation, see DeJong and Spears, 1995) also the co learning algorithm introduced by Shoham and Tennenholtz, 1994). However, the view of innovation as accidental mutation does not do justice to the agents active role in the establishment of conventions. Agents representations and interpretations seem to have a fundamental part in such a phenomenon. 4. This raises the more general problem of the connection ....

Shoham, Y. and Tennenholtz, M. (1994) Co-Learning and the Evolution of Social Activity, CSTR -94-1511, Stanford University.

....by one agent of knowledge about another agent. In either case, the learning related to such processes is a distributed learning process. At the extreme end of distributed learning lie global effects which result as a consequence of local changes at the level of individual agents [Hutchins 1991; Shoham Tennenholtz 1993; Weiss 1993] These effects can be regarded as global learning caused by partial views and feedback at the local level, which nevertheless result in a new consistent behavior of the system in its entirety. Dimensions of Machine Learning in Design 8 D7. Consequences of learning The performance ....

Y. Shoham & M. Tennenholtz, Co-Learning and the Evolution of Social Activity, Tech. Rep. Stanford University, 1993, 36 pp.

....leading to non stationarities. For simplicity, most studies in multi agent learning focus on settings where payo s are either negatively correlated, as in zero sum games (see, for example, Littman [33] or positively correlated, as in coordination games (see, for example, Shoham and Tennenholtz [45]) Two notable exceptions include Wellman and Hu [50] who conducted a theoretical investigation of multi agent learning in market interactions, and Sandholm and Crites [42] who conducted empirical studies of multi agent reinforcement learning in the Prisoners Dilemma. Similarly, this work ....

Y. Shoham and M. Tennenholtz. Co-learning and the evolution of social activity. Mimeo, 1993.

....to learn an (relatively) accurate estimate of the action propensities of other agents. Without such knowledge (and being blind to the choices made by other agents) it may not be possible to reach jointly optimal outcomes (in some sense) as demonstrated by early work in learning games, such as Shoham and Tennenholtz (1994) and Sen and Sekaran (1998) Shoham and Tennenholtz (1994) showed that simple learning rules could be very inecient in cooperative games. Sen and Sekaran (1998) also attempted to learn to coordinate without explicit regard to other agents. They achieved limited success in small domains. Sen and ....

....propensities of other agents. Without such knowledge (and being blind to the choices made by other agents) it may not be possible to reach jointly optimal outcomes (in some sense) as demonstrated by early work in learning games, such as Shoham and Tennenholtz (1994) and Sen and Sekaran (1998) Shoham and Tennenholtz (1994) showed that simple learning rules could be very inecient in cooperative games. Sen and Sekaran (1998) also attempted to learn to coordinate without explicit regard to other agents. They achieved limited success in small domains. Sen and Sekaran (1998) however, pointed out that such learning is ....

[Article contains additional citation context not shown here]

Y. Shoham and M. Tennenholtz, (1994). Co-learning and the evolution of social activity. Technical Report STAN-CS-TR-94-1511, Stanford University.

.... alone, for example, not limited to bidding alone to form a coalition (as in Zlotking and Rosenschein 1994, Ketchpel 1996) or bidding alone as the only means for learning (as in Baum 1997) Neither is it a model of pure reinforcement learning (without explicit interaction among agents, such as Shoham and Tennenholtz 1994, Salustowicz et al. 1998, Claus and Boutilier 1998) But it is the combination and interaction of the two aspects. A dicult issue in RL, which we deal with using the afore described multiagent system, is to deal with non Markovian temporal dependencies (that is, the situations in which actions at ....

Y. Shoham and M. Tennenholtz, (1994). Co-learning and the evolution of social activity. Technical Report STAN-CS-TR-94-1511, Stanford University.

....at run time and requiring programmers to consider all potential interactions at design time. The work reviewed here does not discuss algorithms for generating such laws (although elsewhere Shoham and Tennenholtz investigate systems in which cooperation emerges through the learning of social laws [ST93a, ST92a] Instead, Shoham and Tennenholtz focus on the computational complexity of deriving useful social laws. They first introduce a formalism so that the problem can be stated precisely. They then show that this general problem of deriving a useful social law is NP complete. Finally, they ....

Y. Shoham and M. Tennenholtz. Co-learning and the evolution of social activity. Technical report, Stanford University Department of Computer Science, 1993.

....a certain amount of strength (money) which they exchange for messages at a global blackboard (the market) Much as money mirrors the flow of commodities in a simple commodity economy, value mirrors the flow of messages in the learning classifier system. Most RL algorithms are composed of rules. Shoham Tennenholtz (1994) discuss a generalisation of RL to MAS called co learning. Co learning involves individual agents learning in a social environment that includes other agents. Co learning agents must adapt to each other. Kittock (1995) describes some computational experiments on the emergence of social conventions ....

....TIT FORTAT may form the foundation for certain social emotions, such as gratitude and guilt, which facilitate cooperative behaviour in natural MAS. The existence of a universal norm of reciprocity (Gouldner, 1960) in social behaviour has been known for some time in sociological theory. Shoham Tennenholtz (1994) report experimental results that seem to show that MAS consisting of agents that use a highest cumulative reward rule (i.e. agents that choose actions likely to yield the highest pay off) are inefficient in developing social cooperation. Societies of pure personal utility maximisers may not ....

Shoham, Y. & Tennenholtz, M. (1994). Co-learning and the evolution of social activity. Technical Report CS-TR-94-1511, Robotics Laboratory, Department of Computer Science, Stanford University.

....for an agent society to reach a convention without any centralized control if agents interact and learn from their experiences [ Shoham and Tennenholtz, 1992a ] Conventions thus achieved have been called emergent conventions , and the process for reaching them has been dubbed co learning. Shoham and Tennenholtz, 1993 ] In previous work on the emergence of conventions through co learning, it was assumed that each agent in a society is equally likely to interact with any other agent [ Shoham and Tennenholtz, 1992a ] This seems an unreasonable assumption in the general case, and we consider ways to extend the ....

....we believe that exploring the behavior of societies of simple agents will yield insight into the behavior we can expect from more complex agents. For our preliminary investigations, we have chosen to use a learning rule similar to the Highest Cumulative Reward rule used by Shoham and Tennenholtz [ Shoham and Tennenholtz, 1993 ] To decide which strategy it will use, an agent first computes the cumulative reward for each strategy by summing the feedback from all interactions in which it used that strategy and then chooses the strategy with the highest cumulative reward (HCR) There are, of course, many other possible ....

Yoav Shoham and Moshe Tennenholtz. Co-learning and the evolution of social activity. Submitted for publication, 1993.

....might be designed into agents behavior or legislated by a central authority, emergent conventions are the result of the behavioral decisions of individual agents based on feedback from local interactions. Shoham and Tennenholtz extended this idea into a more general framework, dubbed co learning [6]. In the co learning paradigm, agents acquire experience through interactions with the world, and use that experience to guide their future course of action. A distinguishing characteristic of co learning is that each agent s environment consists (at least in part) of the other agents in the ....

....of co learning is that each agent s environment consists (at least in part) of the other agents in the system. Thus, in order for agents to adapt to their environment, they must adapt to one another s behavior. Here, we describe a modification of the co learning framework as presented in [6] and examine its effects on the emergence of conventions in a model multi agent system. Simulation Model We assume that most tasks that an agent might undertake can only be performed in a limited number of ways; actions are thus chosen from a finite selection of strategies. A convention exists ....

[Article contains additional citation context not shown here]

Yoav Shoham and Moshe Tennenholtz. Co-learning and the evolution of social activity. Submitted for publication, 1993.

....results obtained with this system. 4.3 The Best Choice SR (BCSR) The Best Choice SR (BCSR) is a learning rule that assumes a high value of n, i.e, which always chooses the best resource in a given point. We will assume w is fixed to a given value while discussing BCSR. In our previous work (Shoham Tennenholtz, 1992, 1994), we showed that learning rules that strongly resemble BCSR are useful for several natural multi agent learning settings. This suggests that we need to carefully study it in the case of adaptive load balancing. As we will demonstrate, BCSR is not always useful in the load balancing setting. The ....

....and models (Huberman Hogg, 1988) enable to demonstrate aspects of purely local adaptive behavior in a non trivial model. Our results about the disagreement between selfish interest of agents and the common interest of the population is in sharp contrast to previous work on multi agent learning (Shoham Tennenholtz, 1992, 1994) and to the dynamic programming perspective of earlier work on distributed systems (Bertsekas Tsitsiklis, 1989) Moreover, we explore how the interaction between different agent types affects the system s efficiency as well as Adaptive Load Balancing: A Study in Multi Agent Learning the ....

[Article contains additional citation context not shown here]

Shoham, Y., & Tennenholtz, M. (1994). Co-learning and the evolution of social activity.

No context found.

Y. Shoham and M. Tennenholtz, (1994). Co-learning and the evolution of social activity. Technical Report STAN-CS-TR-94-1511, Stanford University.

No context found.

Shoham, Y.---Tennenholtz, M.: Co-learning and the evolution of social activity. Technical report, Stanford University Department of Computer Science, 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC