P. Dagum, A. Galper, and E. Horvic, "Dynamic network models for forecasting," in Proc. of the 8th ConferenceonUncertainty in Artificial Intelligence, pp. 41--48, 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....almost right from the start at least an initial estimation to base its adaptation decisions on. To address this issue, we embedded the classification procedure described in the previous section in a framework that is able to update its estimates over time based on dynamic Bayesian networks (DBNs, [13]) Essentially, each time a new speech sample becomes available, a new instance of the BN of Figure 3 is connected to the previous ones, the current results of the classifiers are taken into account and the standard BN reasoning algorithms are applied as described before. Figure 4 shows an ....

P. Dagum, A. Galper, and E. Horvitz, "Dynamic network models for forecasting," in Uncertainty in Artificial Intelligence: Proceedings of the Eight Conference. San Francisco: Morgan Kaufman, 1992, pp. 41--48.

....statistical variables; secondly, to model the atemporal relationship between statistical variables. In both cases, the modelling is essentially done in terms of statistical dependence and independence, but the semantics of the two types of relationship are di#erent. The research described in [7, 18, 19, 53, 4] o#ers a starting point for the proposed research. The timebayes project will in particular focus on methods for developing temporal Bayesian network models that take into account the special temporal characteristics of CPR data. Furthermore, methods of sensitivity analysis of temporal Bayesian ....

P. Dagum, A. Galper and E. Horvitz. Dynamic network models for forecasting. In: Proceedings of UAI92, 1992, pp. 41--48.

....of the model, an intelligent forecasting system must update the conditional probabilities, and also the structure if needed, as new evidence arrives. We have previously discussed the effects of unmodeled exogenous forces on DNMs and methods to update the model structure with new evidence [9]. The latter led to the development of unique parametric decompositions, additive decompositions, of the conditional probabilities in the model [11] The theory of combining probability distributions has been extensively investigated by the Bayesian community (e.g. 24] These combination ....

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Proceedings of the 8th Conference on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, Calif., July 1992. American Association for Artificial Intelligence.

....and they should perhaps be renamed as independence networks since what they encode is explicit conditional independencies between variables. Since they were originally proposed, the use of belief networks has become widespread. There are numerous applications that make use of them, for example [2, 6, 29, 99, 108]. They have even been proposed as a means of establishing the best document to retrieve from a document database [155] an information IEEE Transactions on Knowledge and Data Engineering, 8(3) 353 372. 25 retrieval application that clearly has connections with the needs of database users. There ....

....diagnosed changes over time, and so the history of the problem becomes important. This time dependency is handled in Provan s system Dynasty, which also allows di erent levels of granularity of problem description to be considered during the diganosis. Similar issues are addressed by Dagum et al. [29] who synthesize belief networks with time series analysis to create dynamic network models for use in forecasting. Another factor that has been disregarded in all the systems considered so far is the problem of seperating model construction from evaluation. In a resource bound environment this ....

Dagum, P., Galper, A., and Horvitz, E. (1992) Dynamic network models for forecasting, Proceedings of the 8th Conference on Uncertainty in Articial Intelligence, Stanford, 41-48.

....or time element into the model. This has been approached in various different ways. Aliferis and Cooper [1] summarise just some of the extensions of belief networks for time modeling presented over the last few years. These include temporal influence diagrams [33] Dynamic Belief Networks (DBNs) [7], temporal models of endogenous change [13] Temporal Bayesian Networks (TBNs) 40] Temporal Nodes Bayesian Network (TNBNs) 3] embedded Markov processes [4] logic and time nets [19] 20] Modifiable Temporal Belief Networks (MTBNs) 1] as well as specific applications [4] 30] An obvious ....

....slice is used to represent a snapshot of the evolving temporal process [21] The belief network consists of a sequence of sub models each representing the system at a particular point or interval in time (time slice) and which are interconnected by temporal relations. Kjaerulff [22] Dagum et al.[7][8] Provan[32] Berzuini[4] Lekuona[26] are just some of the researchers currently using the time slice approach. a) b) Figure 2: Diagram showing the difference between models based on time slices (adapted from Hanks [13] Figure 2(b) represent models where the network is composed of ....

P. Dagum, A. Galper, E. Horvitz. "Dynamic Network Models for Forecasting", Proc. of the 8 th Workshop on Uncertainty in Artificial Intelligence, pp.41-48, 1992.

....non interactive and often incomplete observations about a user as looking into a room through a keyhole . The system first must learn which actions and positions or sequences of actions and positions tend to lead to a particular quest. This information is modeled in a dynamic Bayesian network [17]. During the testing phase, the dynamic Bayesian network is used to predict the player s quest, next action, and next location. 2.4 Examples of Adaptive Hypermedia Systems 20 2.4.9 POKS POKS [22] is based on a cognitive theory of knowledge structures. It builds a network of implications on ....

Dagum, P., Galper, A., and Horvitz, E. Dynamic network models for forecasting. In Eighth Conference on Uncertainty in Artificial Intelligence (San Mateo, 1992), Morgan Kaufmann Publishers, Inc., pp. 41--48.

....size, is NP hard w.r.t. the accuracy of the estimated probabilities [7] Under the light of these theoretical results, the inexact inference methods can be useful in large networks where a certain degree of error in the estimated probabilities is tolerable. Dynamic Probabilistic Network (DPN) [10, 23, 8] is a special Bayesian network architecture developed for modelling a dynamic environment. A DPN is made up from a number of time slices where each time slice represents the state of the environment at the current time. Each time slice can be a Bayesian network in its own. The dynamics of the ....

Paul Dagum, Adam Galper, and Eric Horvitz. Dynamic network models for forecasting. In Proceedings of the Eighth Annual Conference on Uncertainty in Articial Intelligence, pages 41-48, 1992.

....structures and non linear relationships of have appeared to be rather modest. By formulating the analysis in terms of DPNs both of these limitations vanish. Attempts to integrate methods of classical time series analysis with network representation and inference techniques have been presented by Dagum, Galper Horvitz (1992). This paper, however, does not address the issue of model assessment, but merely problems related to making inferences (including prediction and backward smoothing, in classical time series analysis terms) That is, the dynamic model is assumed to be given. Among research activities applying ....

Dagum, P., Galper, A. & Horvitz, E. (1992). Dynamic network models for forecasting, Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence, Morgan Kaufmann Publishers, Inc., pp. 41--48.

....paper, we address three practical modeling issues. 1 Introduction Bayesian networks (BNs) Pea88; Nea90; Jen96) provide a normative formalism for diagnosis based on probabilistic domain knowledge. In the past decade, researchers have studied how to model diagnostic problems using BNs (Hec90; DGH92; HBR95) and many algorithms have been proposed to perform inference in BNs (Pea88; Sha96; CGH97; Jen96) Most of these methods are based on a flat BN representation of the system to be diagnosed. As BNs become widely accepted, they are applied to larger and more complex problem domains. ....

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In D. Dubois, M.P. Wellman, B. D'Ambrosio, and P. Smets, editors, Proc. 8th Conf. on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, CA, 1992.

....the robot remain within tolerable limits on certainty. Opportunistic reduction of sensing is a challenging problem for several inter related reasons. First, it requires some mechanism for projecting the decay in sensing certainty over time. Existing mechanisms such as dynamic belief networks [ Dagum et al. 1992; Dean and Kanazawa, 1989 ] and survivor functions [ Dean and Wellman, 1991 ] are unsatisfactory for autonomous mobile robots. These mechanisms assume the existence of explicit a priori models of the object s behavior, interactions with the environment, and possible contravening events in order ....

P. Dagum, A. Glaper, and E. Horvitz. Dynamic network models for forecasting. In Eighth Conference on Uncertainty in Artificial Intelligence, pages 41--48, 1992.

....the instantaneous belief must be applied. However, such a method must take into account contradictory influences; for example, older belief tends to be less believable [ Drainkov and Lang, 1993 ] but objects tend to persist [ Dean and Wellman, 1991 ] Methods such as dynamic belief networks [ Dagum et al. 1992; Dean and Kanazawa, 1989 ] and survivor functions [ Dean and Wellman, 1991 ] are unsatisfactory for autonomous mobile robots operating in unknown domains, both in theory and in practice. These methods assume explicit a priori models of the object s behavior, interactions with the environment, and ....

P. Dagum, A. Glaper, E. Horvitz. Dynamic network models for forecasting. In Eighth Conference on Uncertainty in Artificial Intelligence, pages 41--48, 1992.

....time slice networks (which may be modified dynamically by some external agent) before entered into the dHugin system. Attempts to integrate methods of classical time series analysis with Bayesian network representation and inference techniques have been presented by Kenley (1986) and Dagum et al. [Dagum et al. 1992), Dagum Galper (1993a) Dagum Galper (1993b) The element from classical time series analysis represented in the method of Dagum et al. concerns estimation of a single parameter allowing the dynamic model to adapt to unexpected changes in the time series. The parameter controls the relative ....

Dagum, P., A. Galper and E. Horvitz, 1992, "Dynamic network models for forecasting", Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence (Morgan Kaufmann Publishers, San Mateo, California), 41--48.

....Murphy s rule is O(c) since the belief has only two propositions, for and against and only two beliefs are combined on each update. Furthermore, survivor functions require global reasoning about the actions of the robot and contravening events. Another related mechanism is dynamic network models [ Dagum et al. 1992 ] The values for propagating belief are encapsulated in a static belief network, but change dynamically based on the probabilistic time series analysis in [ West and Harrison, 1989 ] This method also has the same limitations as temporal probabilistic networks. The real advantage of direct ....

P. Dagum, A. Glaper, and E. Horvitz. Dynamic network models for forecasting. In Eighth Conference on Uncertainty in Artificial Intelligence, pages 41--48, 1992.

.... such as the fall diagnosis problem, where the world changes and the focus is reasoning over time [7, 11, 15, 10] Such dynamic applications include robot navigation and map learning based on temporal belief networks [7] monitoring diabetes [1] monitoring robot vehicles [13] oil forecasting [5], 17] forecasting sleep apnea [4] automated vehicle control [8] and traffic plan recognition [18] For such applications the network grows over time, as the state of each domain variable at different times is represented by a series of nodes. These dynamic networks are Markovian, which ....

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Proceedings of the 8th Conference on Uncertainty in Artificial Intelligence, pages 41--48, 1992.

....physics problem. This BN is then used to identify a student s problem solving strategy and predict his or her next step. Dynamic applications are characterized by a constantly changing world. In order to model this change, temporal reasoning must be incorporated into BNs (Dean and Wellman, 1991; Dagum et al. 1992; Nicholson and Brady, 1994) This is done by allowing a BN to grow over time, and representing the state of each domain variable at different times by a series of nodes. Typically, for these Dynamic Belief Networks (DBNs) the connections over time are Markovian, and a temporal window is ....

Dagum, P., Galper, A., and Horvitz, E. (1992). Dynamic network models for forecasting. In UAI92 -- Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, California.

....of a monitoring task, showing that orders of magnitude faster inference can be achieved with only a small degradation in accuracy. page 1 of 1 Introduction The ability to model and reason about stochastic processes is fundamental to many applications [ Forbes et al. 1995; Jensen et al. 1989; Dagum et al. 1992; Provan, 1992 ] For example, we may be observing a freeway traffic scene via a video camera mounted on a bridge, with the goal of understanding the current traffic situation and predicting its future evolution [ Forbes et al. 1995 ] or monitoring a patient s symptoms to design his treatment ....

....state variables, and a compact representation of the probabilistic model by utilizing conditional independence assumptions. Here, a belief state is a distribution over some subset of the state variables at time t. In general, not all of the variables at time t must participate in the belief state [ Dagum et al. 1992 ] however, at least) every variable whose value at time t directly affects its value at time t 1 must be included. In large DBNs, the obvious representation of a belief state (as a flat distribution over its state space) is therefore typically infeasible, particularly in time critical ....

P. Dagum, A. Galper, and E. Horwitz. Dynamic network models for forecasting. In Proc. UAI, 1992.

....they do not consider stochastic processes for modeling the temporal evolution. Kjaerulff [Kjaerulff92] proposes a schema for reasoning in dynamic probabilistic networks and he consideres Markov chains for modeling the evolution of the system. Dagum and Galper [Dagum93a] Dagum, Galper and Horvitz [Dagum92] and Dagum, Galper, Horvitz and Seiver [Dagum93b] simplify the assessment of conditional probabilities in dynamic networks by using simple parametric decompositions. Whenever certain dependence conditions do not hold such assumptions are not easily justifiable. 76 Alberto Lekuona, Beatriz Lacruz ....

Dagum, P.; Galper, A.; Horvitz, E. (1992) Dynamic network models for forecasting. In Dubois, D.; Wellamn, M.P.; D'Ambrosio, B.; Smets, P. (Eds.) Uncertainty in Artificial Intelligence. Proceedings of the Eighth Conference. Morgan Kaufmann, 41-48.

....structures and non linear relationships of have appeared to be rather modest. By formulating the analysis in terms of DPNs both of these limitations vanish. Attempts to integrate methods of classical time series analysis with network representation and inference techniques have been presented by Dagum, Galper and Horvitz (1992). This paper, however, does not address the issue of model assessment, but merely problems related to making inferences (including prediction and backward smoothing, in classical time series analysis terms) That is, the dynamic model is assumed to be given. Among research activities applying ....

Dagum, P., Galper, A. and Horvitz, E. (1992). Dynamic network models for forecasting, Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence.

No context found.

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Proceedings of the Eighth Workshop on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, CA, July 1992. Association for Uncertainty in Artificial Intelligence.

No context found.

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, CA, July 1992. Association for Uncertainty in Artificial Intelligence, Morgan Kaufmann.

No context found.

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Proceedings of the Eighth Workshop on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, CA, July 1992. Association for Uncertainty in Artificial Intelligence.

No context found.

P. Dagum, A. Galper, and E. Horvic, "Dynamic network models for forecasting," in Proc. of the 8th ConferenceonUncertainty in Artificial Intelligence, pp. 41--48, 1992.

No context found.

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, July 1992.

No context found.

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In Didier Dubois, Michael P. Wellman, Bruce D'Ambrosio, and Phil Smets, editors, Uncertainty in Artificial Intelligence: Proceedings of the Eighth Conference, pages 41-- 48, San Francisco, 1992. Morgan Kaufmann.

No context found.

P. Dagum, A. Galper, and E. Horvitz. Dynamic network models for forecasting. In D. Dubois, M.P. Wellman, B. D'Ambrosio, and P. Smets, editors, Proc. 8th Conf. on Uncertainty in Artificial Intelligence, pages 41--48, Stanford, CA, 1992.

First 50 documents

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