M. Ramoni, "Ignorant influence diagrams", in Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence (IJCAI-95), pp. 1808-1814, S. Mateo, CA, 1995. Morgan Kaufman.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....probability intervals rather than point valued probabilities computed under a specific model for the missing data. The second feature of the RBC is its ability to classify cases by reasoning with probability in tervals. Interval probabilities have been studied by several authors, see for example [5, 6]. The interval based classification in an RBC is based on a propagation algorithm that computes posterior probability intervals containing all the scoring values that could be obtained from the exact computation of all possible completions of the training set. However, since the result of this ....

M. Ramoni, "Ignorant influence diagrams", in Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence (IJCAI-95), pp. 1808-1814, S. Mateo, CA, 1995. Morgan Kaufman.

....(Doan 1996) Such abstract worlds cannot be represented in the SPD framework. Most of the work on computing with probability intervals has been done in the context of probabilistic logic, Bayesian networks, and influence diagrams (Nilsson 1986; Frisch Haddawy 1994; Breese Fertig 1991; Ramoni 1995). There has been little work dealing directly with the restrictiveness of the single probability distribution model in planning. An exception is the work of Chrisman (Chrisman 1992) who develops an action model based on the notion of DempsterShafer mass functions. His work concentrates on ....

....of view of asymptotic complexity, finiteness is sufficient. is infeasible in practice in all but small domains. Most of the work in extending the restrictiveness of the single probability distribution have been placed in the context of belief network and influence diagram (Breese Fertig 1991; Ramoni 1995). Some deal directly with building action models and rules for projecting actions (Chrisman 1992) The restrictiveness of these models is that they all assume a finite state space; the inference algorithm (marginalization in belief net and forward projection in action model) takes advantage of ....

Ramoni, M. 1995. Ignorant influence diagrams. In Proceedings of the International Joint Conference on Artificial Intelligence, 1869--1875.

....inferencing is possible. 3 Often, these numerous (combinatoric) probabilities are not known, and do not make any sense to a human; defining them becomes a major stumbling block in knowledge acquisition. 3 There is some current work on inferencing without a complete probability specification, [38, 39]; however, this work does not accommodate finer knowledge resolution nor allow for incomplete RV state specification. This approach, while valid, is orthogonal to the work presented here which directly addresses those limitations of Bayesian Networks. See Section 3.3) 3 3 R=false S=dry R=true ....

Ramoni, Marco. "Ignorant Influence Diagrams." IJCAI-95: Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence. 1869--1875. 1995.

....as a set of constraints on the possible distributions in the database. In this way, our learning algorithm returns probability intervals which account for the reliability of the information available in the database. These intervals can be then propagated using current techniques, such as [5]. An experimental comparison between our method and a stochastic method shows a remarkable difference in accuracy between the two methods and the computational advantages of our deterministic method with respect to the stochastic one. Acknowledgments Authors thank Greg Cooper, Pat Langley, Paul ....

M. Ramoni. Ignorant influence diagrams. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1808--1814, S. Mateo, CA, 1995. Morgan Kaufman.

....since ignorance is explicitly represented in the ibn, it can be taken into account in the development of the reasoning process. Since bbns naturally extend into a decision method called Influence Diagrams, we are developing a class of Influence Diagrams able to make decisions under ignorance [30]. In this way, the blood glucose forecaster will be able to suggest adjustments of insulin dosage in response to the predicted blood glucose concentration. Acknowledgments This research was supported in part by the AIM Programme of the Commission of the European Communities (A2034) and by a grant ....

M. Ramoni, Ignorant influence diagrams, in: Proceedings of the International Joint Conference on Artificial Intelligence (1995).

.... a system to forecast blood glucose concentration in insuling dependent diabetic patiens, using the probabilistic information direclty extracted from a clinical database [17] ibns have been recently extended into a complete decision theoretic formalism called Ignorant Influence Diagrams (iids) [13]. iids implement an inference policy, largely wished for in the literature about probabilistic reasoning systems, called incremental refinement policy [4] able to improve the accuracy of the solutions as a monotonically increasing function of the allocated resources and the available information. ....

M. Ramoni. Ignorant influence diagrams. In Proceedings of the International Joint Conference on Artificial Intelligence, 1995. Forthcoming.

.... of the computational advantages provided by conditional independence assumptions with the expressive power of probability intervals [van der Gaag, 1991] Therefore, some efforts have been addressed to extend bbns from real valued probabilities to interval probabilities [Breeze and Fertig, 1990, Ramoni, 1995] thus combining the advantages of conditional independence assumptions with the explicit representation of ignorance. These efforts provide different methods to use the bbns we can learn with our method. 3.2 Learning We will describe the method we propose by using the artificial database in ....

....in X . 4.2 Storing It is apparent that the procedure store plays a crucial role for the efficiency of the procedure. In order to develop an efficient algorithm, we used discrimination trees to store the parameter counters, following a slightly modified version of the approach proposed by [Ramoni et al. 1995]. Along this approach, each state of each variable of the network is assigned to a discrimination tree. Each level of the discrimination tree is defined by the possible states of a parent variable. Each path in the discrimination tree represents a possible configuration of parent variables for ....

[Article contains additional citation context not shown here]

M. Ramoni. Ignorant influence diagrams. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1808-- 1814, S. Mateo, CA, 1995. Morgan Kaufman.

No context found.

M. Ramoni. Ignorant influence diagrams. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1869--1875, Montreal, Canada, August 1995.

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