8. 5 Summary In this chapter we have identified three basic patterns of influences that lead to ambiguity in the QP analysis of the basic active furnace state. We have then shown how modification of these patterns, by adding equilibrium values and sensitivity annotations on influence arcs, could permit resolu­ tion of the ambiguities. Finally, we have described in detail the extensions needed to the basic influence resolution algorithm in QP theory to oper­ ate on these extended descriptions. We have also shown that the modified influence resolution algorithm corrects an error in Forbus'' original method for combining influences. We have then presented an extended example in which introduction of equilibrium assumptions eliminates all ambigu­ ity in the influence resolution deduction. In the next chapter we extend these techniques further, by developing a qualitative perturbation analysis technique that permits us to answer "what ir'' control questions; then we extend this technique to obtain quantitative, as well as qualitative, effects of hypothetical control actions. 8.

8. 5 Summary In this chapter we have identified three basic patterns of influences that lead to ambiguity in the QP analysis of the basic active furnace state. We have then shown how modification of these patterns, by adding equilibrium values and sensitivity annotations on influence arcs, could permit resolu­ tion of the ambiguities. Finally, we have described in detail the extensions needed to the basic influence resolution algorithm in QP theory to oper­ ate on these extended descriptions. We have also shown that the modified influence resolution algorithm corrects an error in Forbus' original method for combining influences. We have then presented an extended example in which introduction of equilibrium assumptions eliminates all ambigu­ ity in the influence resolution deduction. In the next chapter we extend these techniques further, by developing a qualitative perturbation analysis technique that permits us to answer "what ir' control questions; then we extend this technique to obtain quantitative, as well as qualitative, effects of hypothetical control actions. 8.

Uncertainty in Artificial Intelligence: Proceedings of the Eighth Conference (1992) covers the papers presented at the Eighth Conference on Uncertainty in Artificial Intelligence, held at Stanford University on July 17-19, 1992. The book focuses on the processes, methodologies, technologies, and approaches involved in artificial intelligence. The selection first offers information on Relative Evidential Support (RES), modal logics for qualitative possibility and beliefs, and optimizing causal orderings for generating DAGs from data. Discussions focus on reversal, swap, and unclique operators, modal representation of possibility, and beliefs and conditionals. The text then examines structural controllability and observability in influence diagrams, lattice-based graded logic, and dynamic network models for forecasting. The manuscript takes a look at reformulating inference problems through selective conditioning, entropy and belief networks, parallelizing probabilistic inference, and a symbolic approach to reasoning with linguistic quantifiers. The text also ponders on sidestepping the triangulation problem in Bayesian net computations; exploring localization in Bayesian networks for large expert systems; and expressing relational and temporal knowledge in visual probabilistic networks. The selection is a valuable reference for researchers interested in artificial intelligence.

8. 5 Summary In this chapter we have identified three basic patterns of influences that lead to ambiguity in the QP analysis of the basic active furnace state. We have then shown how modification of these patterns, by adding equilibrium values and sensitivity annotations on influence arcs, could permit resolu­ tion of the ambiguities. Finally, we have described in detail the extensions needed to the basic influence resolution algorithm in QP theory to oper­ ate on these extended descriptions. We have also shown that the modified influence resolution algorithm corrects an error in Forbus'' original method for combining influences. We have then presented an extended example in which introduction of equilibrium assumptions eliminates all ambigu­ ity in the influence resolution deduction. In the next chapter we extend these techniques further, by developing a qualitative perturbation analysis technique that permits us to answer "what ir'' control questions; then we extend this technique to obtain quantitative, as well as qualitative, effects of hypothetical control actions. 8.

8. 5 Summary In this chapter we have identified three basic patterns of influences that lead to ambiguity in the QP analysis of the basic active furnace state. We have then shown how modification of these patterns, by adding equilibrium values and sensitivity annotations on influence arcs, could permit resolu­ tion of the ambiguities. Finally, we have described in detail the extensions needed to the basic influence resolution algorithm in QP theory to oper­ ate on these extended descriptions. We have also shown that the modified influence resolution algorithm corrects an error in Forbus' original method for combining influences. We have then presented an extended example in which introduction of equilibrium assumptions eliminates all ambigu­ ity in the influence resolution deduction. In the next chapter we extend these techniques further, by developing a qualitative perturbation analysis technique that permits us to answer "what ir' control questions; then we extend this technique to obtain quantitative, as well as qualitative, effects of hypothetical control actions. 8.

Uncertainty in Artificial Intelligence: Proceedings of the Eighth Conference (1992) covers the papers presented at the Eighth Conference on Uncertainty in Artificial Intelligence, held at Stanford University on July 17-19, 1992. The book focuses on the processes, methodologies, technologies, and approaches involved in artificial intelligence. The selection first offers information on Relative Evidential Support (RES), modal logics for qualitative possibility and beliefs, and optimizing causal orderings for generating DAGs from data. Discussions focus on reversal, swap, and unclique operators, modal representation of possibility, and beliefs and conditionals. The text then examines structural controllability and observability in influence diagrams, lattice-based graded logic, and dynamic network models for forecasting. The manuscript takes a look at reformulating inference problems through selective conditioning, entropy and belief networks, parallelizing probabilistic inference, and a symbolic approach to reasoning with linguistic quantifiers. The text also ponders on sidestepping the triangulation problem in Bayesian net computations; exploring localization in Bayesian networks for large expert systems; and expressing relational and temporal knowledge in visual probabilistic networks. The selection is a valuable reference for researchers interested in artificial intelligence.