T. Sugawara and V. Lesser. On-line learning of coordination plans. In Working Papers of the 12th International Workshop on Distributed Artificial Intelligence, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....coordination through joint planning (e.g. 2, 3, 6, 8, 10, 11, 17, 20] However, there are only very little approaches that combine joint learning and joint planning. There are two exceptions that are related to the JPJL algorithm. The first is the work by Sugawara and Lesser described in e.g. [21, 22]. The basic idea behind this approach is to enable agents to learn situation specific rules that capture relevant non local information in order to improve local planning and reasoning. This idea has been investigated within the context of LODES, a distributed diagnosis system for computer ....

T. Sugawara and V. Lesser, `On-line learning of coordination plans', in Working Papers of the 12th International Workshop on Distributed Artificial Intelligence, (1993).

.... M Dyna Q approach unique and different from a number of related approaches to multiagent activity coordination, including approaches that rely on either pure planning or pure reaction (see the references provided in Section 1) approaches that rely on a combination of planning and learning (e.g. [31, 32, 40]) and approaches that rely on a combination of reacting and learning (e.g. 2, 18, 21, 23, 28, 26, 38, 39] Obviously M Dyna Q can be considered as a generalization of these approaches, and as such it o ers maximum coordination exibility. This is not to say that M Dyna Q is always the best ....

T. Sugawara and V. Lesser. On-line learning of coordination plans. In Working Papers of the 12th International Workshop on Distributed Articial Intelligence, 1993.

....(e.g. DL92, DKK97, DL91, Geo83, HD96, Kab95, SH96, Sug95, von92] However, there are only very little approaches that combine joint learning and joint planning. There are two exceptions that are related to the JPJL algorithm. The rst is the work by Sugawara and Lesser described in e.g. SL93, SL98] The basic idea behind this approach is to enable agents to learn situation speci c rules that capture relevant non local information in order to improve local planning and reasoning. This idea has been investigated within the context of LODES, a distributed diagnosis system for computer ....

T. Sugawara and V. Lesser. On-line learning of coordination plans. In Working Papers of the 12th International Workshop on Distributed Articial Intelligence, 1993.

....of machine learning and the study of game playing have come together to yield several significant advances. To date, research in multiple agent planning and control has been limited largely to the area of distributed artificial intelligence (Rosenschein and Genesereth 1985; Stone and Veloso 1996a; Suguwara and Lesser 1993; Tan 1993) and artificial life (Collins 1992; Huberman and Glance 1995; Sandholm and Crites 1995; Stanley, Ashlock, and Tesfastsion 1993) In distributed AI (DAI) several agents cooperate to achieve some goal or accomplish some task. The task is usually one of sufficient complexity that no ....

Suguwara, T. and Lesser, V. (1993). On-line learning of coordination plans. Technical Report COINS TR 93-27, University of Massachusetts, Amherts, Massachusetts.

....of eventual cooperation, in practice, agents using this learning rule usually fail to reach cooperation in hundreds of thousands of iterations. Other multiagent learning research has used purely heuristic algorithms for complex real world problems such as learning coordination strategies [41] and communication strategies [16] with varying success. Our work differs from the above mentioned works since it uses a reinforcement learning approach to make meta level control decisions in a complex cooperative environment. None of these works reason about the cost of reasoning and there is no ....

Toshiharu Sugawara and Victor Lesser. On-line learning of coordination plans. In Proceedings of the 12th International Workshop on Distributed Artificial Intelligence, pages 335-- 345,371--377, 1993.

....they know the existence of the narrow bandwidth line. The existence of this problem and the fact that it was caused by the agents themselves will be the trigger for the learning component to be invoked. Other mechanisms for invoking the learning component are discussed by Sugawara and Lesser [22]. 3. The Learning Framework 3.1. Behaviors of Planner and Scheduler Before presenting our learning framework, it is necessary to describe the LODES agent control framework, especially the behaviors of the planning and scheduling components. During problem solving, the planner of an agent ....

....other agents to decide on which tasks to locally select and their priorities. Because of space limitations, we will not discuss how this action part is chosen in this situation from a library of pre existing explicit coordination strategies, for more details see the work by Sugawara and Lesser [22]. a) L1 s subjective viewtype: type: scarceL1L2L1L3L7nam ed Get RTT betweenBoth Endsrule and the plan GetRTT between BothEnds is selected in more than two agents ; that is, the view of the current situation is identical to (b) Min of MaxThrput ; that is, the type of ....

[Article contains additional citation context not shown here]

Sugawara, T. and Lesser, V. R. "On-Line Learning of Coordination Plans," COINS Technical Report, 93-27, Univ. of Massachusetts, 1993. (The shorter version of this paper is also published in Proc. of the 12th Int. Workshop on Distributed AI, 1993.)

....Whinston[10] discuss a classifier system based multi agent learning system that represents an interesting alternative to the learning mechanism we proposed in this paper but as discussed previously, in this paper our interest is more than just learning the organizational roles. Sugawra and Lesser[15] discuss a distributed networking system where each local segment of the network has an intelligent diagnosis agent called LODES that monitors traffic on the network and uses an explanation based learning technique to develop coordination rules for the LODES agents. TEAM like systems may not be ....

T. Sugawara, and V. R. Lesser, "On-Line Learning of Coordination Plans," Twelfth Annual Workshop on Distributed Artificial Intelligence, 1993.

....alternative to the learning mechanism we proposed in this paper for learning organizational knowledge. Nagendra Prasad et al. Nagendra Prasad, Lesser, Lander 1996) and Garland and Alterman(Garland Alterman 1996) discuss issues in knowledge reuse in multi agent systems. Sugawra and Lesser(Sugawara Lesser 1993) discuss a distributed network diagnosis system where each local segment of the network has an intelligent diagnosis agent called LODES that monitors traffic on the network and uses an explanation based learning technique to develop coordination rules for the LODES agents. Unlike these systems, ....

Sugawara, T., and Lesser, V. R. 1993. On-line learning of coordination plans. In Proceedings of the Twelfth International Workshop on Distributed AI.

....process based on this model. Two cases are possible depending on the result of the 2 This is done by the meta level controller within an agent or by an external monitor for specific (shared) resources (such as network resource or database) What kinds of events are detected are described in [17]. It is also assumed that an agent locally records an abstracted trace of its recent problem solving actions which can be reviewed on line by the learning component. reproduction. The first case, which we will call a lack of information problem, can be solved in a very local way by just choosing ....

....however, since L1 . L7 diagnose the same secondary problem in this example, agents can identify the situation by comparing their traces 3 This approach assumes that we have stored some amount of history in the agent about past problem solving experiences in a way to facilitate such analysis [17]. Table 1 (a) Comparative Analysis (L5 and L6) Variables values in L5 values in L6 Adjacent networks Net4, Net6 Net5 and Net7 eliminated Adjacent routers R4, R5 R6, R7 eliminated End nodes (L1, L7) L1, L7) Src MAC xx:xx:xx:f:2a:3b xx:xx:xx:0:12:8c eliminated Dst MAC xx:xx:xx:f:2a:3a ....

[Article contains additional citation context not shown here]

Sugawara, T. and Lesser, V. R. "On-Line Learning of Coordination Plans," COINS Technical Report, 93-27, Univ. of Massachusetts, 1993. (The shorter version of this paper is also published in Proc. of the 12th Int. Workshop on Distributed AI, 1993.)

....strategies to acquire non local views of task structures may not be worthwhile for all problem solving situations. In certain situations, coordination protocols that permit some level of non coherent activity and avoid the additional overhead for coordination may lead to better performance[7, 6, 23, 34]. For example, when the agents are under severe time pressure and the load of the activities at the agents is high, sophisticated agent coordination strategies do not generally payoff. Agents may not have the time to benefit from the increased awareness they derive through coordination in such ....

....process for cooperatively constructing an overall solution. The system was tested in a parametric design domain and the learning agents produced designs that, on an average, were better than those produced by a system with agents playing roles hand coded by a human expert. Sugawara and Lesser[34] also recognize the need for situation specificity in learning coordination, though they do have the notion of two phase coordination. They are concerned with learning to make the situations more discriminating to avoid using an inappropriate coordination strategy in the domain of distributed ....

[Article contains additional citation context not shown here]

T. Sugawara and V. R. Lesser. On-line learning of coordination plans. In Proceedings of the Twelfth International Workshop on Distributed AI, Hidden Valley, Pa, May 1993.

....control knowledge in a more global abstract situation. The information shared is weak and they are studied in domains such as predator prey[Tan, 1993] or blocks world [Weiss, 1994] where the need for sharing meta level information and situating learning in it is not apparent. Sugawara and Lesser[Sugawara Lesser, 1993] also recognize the need for situation specificity in learning coordination, though they do have the notion of two phase coordination. They are concerned with learning to make the situations more discriminating to avoid using an inappropriate coordination strategy in the domain of distributed ....

....of coordination alternatives become large in number, the learning phase could become computationally very intensive and the instance base size could increase enormously with respect to Mode 2. We are looking at how to integrate methods for progressively refining situation vectors such as those in [Sugawara Lesser, 1993] , ways to organize the instance base to access and detect regions where there is insufficient learning and also ways to do more directed experimentation during learning rather than randomly sampling the problem space. In COLLAGE, all the agents form identical instance bases. We could as well have ....

Sugawara, T., and Lesser, V. R. 1993. On-line learning of coordination plans. In Proceedings of the Twelfth International Workshop on Distributed AI.

....alternative to the learning mechanism we proposed in this paper for learning organizational knowledge. Nagendra Prasad, Lesser and Lander(Nagendra Prasad et al. 1996) and Garland and Alterman (Garland and Alterman, 1996) discuss issues in knowledge reuse in multi agent systems. Sugawra and Lesser(Sugawara and Lesser, 1993) discuss a distributed networking system where each local segment of the network has an intelligent diagnosis agent called LODES that monitors traffic on the network and uses an explanation based learning technique to develop coordination rules for the LODES agents. Unlike these systems, TEAM like ....

Sugawara, T. and Lesser, V. R. (1993). On-line learning of coordination plans. In Proceedings of the Twelfth International Workshop on Distributed AI, Hidden Valley, Pa.

....cooperative control. As mentioned previously, this thesis deals with agents characterized by much higher sophistication and heterogeneity. In addition, the problem solving and learning among agents are characterized by explicit cooperation through direct communication. 26 Sugawara and Lesser[Sugawara and Lesser, 1993] discuss a self diagnosing Local Area Networking(LAN) system where each node or a network segment has an intelligent agent called LODES that monitors traffic on the LAN. Initially the agents rely on coarse control strategies for this purpose. Upon detecting problems with these control rules, the ....

....for all problem solving situations. In certain situations, coordination protocols that permit some level of non coherent activity and avoid the additional overhead for coordination may lead to better performance[Durfee and Lesser, 1988, Decker and Lesser, 1995, Nagendra Prasad et al. 1996a, Sugawara and Lesser, 1993] For example, when the agents are 71 under severe time pressure and the load of the activities at the agents is high, sophisticated agent coordination strategies do not generally payoff. Agents may not have the flexibility in their own problem solving, given the time constraints, to benefit ....

[Article contains additional citation context not shown here]

Sugawara, T. and Lesser, V. R. On-line learning of coordination plans. In Proceedings of the Twelfth International Workshop on Distributed AI, Hidden Valley, Pa, May 1993.

No context found.

T. Sugawara and V. Lesser. On-line learning of coordination plans. In Working Papers of the 12th International Workshop on Distributed Artificial Intelligence, 1993.

No context found.

T. Sugawara and V. Lesser, `On-line learning of coordination plans', in Working Papers of the 12th International Workshop on Distributed Artificial Intelligence, (1993).

No context found.

Sugawara, T. and Lesser, V. 1993. On-Line Learning of Coordination Plans. Comp. Sci. TR-93-27, Univ. of Mass., Amherst.

No context found.

Sugawara, T. and Lesser, V. 1993. On-Line Learning of Coordination Plans. Computer Science Technical Report 93-27, University of Massachusetts, Amherst.

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