Tradeoffs in Knowledge-Based Construction of Probabilistic Models

In many domains, the ability to use a knowledge base to automatically construct alternative probabilistic network models and then compare them is desirable. This paper makes two novel contributions towards achieving that goal: first, it analyzes a parameterized class of (a) static, and (b) temporal influence diagram, models which differ in the time-series process describing the temporal evolution of the system being modeled. Second, it applies general scoring metrics for comparing these models with respect to predictive accuracy and computational efficiency. The network rankings facilitate comparing the accuracy/efficiency tradeoffs entailed in using TIDs which differ in (1) the accuracy of capturing the temporal evolution of a dynamic system and (2) data and computational requirements. The scoring metrics are used to compare networks in which all variables evolve according to a Markov process with two novel domain-dependent network approximations. These approximations model the evolution of a parsimonious subset of variables rather than all variables. >

[1]  Keiji Kanazawa,et al.  A model for reasoning about persistence and causation , 1989 .

[2]  Michael P. Wellman,et al.  Planning and Control , 1991 .

[3]  Uffe Kjærulff,et al.  A Computational Scheme for Reasoning in Dynamic Probabilistic Networks , 1992, UAI.

[4]  Gregory F. Cooper,et al.  The Computational Complexity of Probabilistic Inference Using Bayesian Belief Networks , 1990, Artif. Intell..

[5]  Robert P. Goldman,et al.  Integrating Model Construction and Evaluation , 1992, UAI.

[6]  Eric Horvitz,et al.  Reasoning about the Value of Decision-Model Refinement: Methods and Application , 1993, UAI.

[7]  Paul Dagum,et al.  Additive Belief-Network Models , 1993, UAI.

[8]  Ewart R. Carson,et al.  A Model-Based Approach to Insulin Adjustment , 1991, AIME.

[9]  H. Akaike.  A new look at the statistical model identification , 1974 .

[10]  Frank Jensen,et al.  Approximations in Bayesian Belief Universe for Knowledge Based Systems , 2013, UAI 1990.

[11]  Gregory M. Provan,et al.  Dynamic Network Construction and Updating Techniques for the Diagnosis of Acute Abdominal Pain , 1993, IEEE Trans. Pattern Anal. Mach. Intell..

[12]  Henri Jacques Suermondt,et al.  Explanation in Bayesian belief networks , 1992 .

[13]  R. A. Leibler,et al.  On Information and Sufficiency , 1951 .

[14]  Eric Horvitz,et al.  Ideal reformulation of belief networks , 1990, UAI.

[15]  Dean P. Foster,et al.  The risk inflation criterion for multiple regression , 1994 .

[16]  Eric Horvitz,et al.  Dynamic Network Models for Forecasting , 1992, UAI.

[17]  Keiji Kanazawa,et al.  A Logic and Time Nets for Probabilistic Inference , 1991, AAAI.

[18]  G. Schwarz.  Estimating the Dimension of a Model , 1978 .

[19]  David J. Spiegelhalter,et al.  Sequential Model Criticism in Probabilistic Expert Systems , 1993, IEEE Trans. Pattern Anal. Mach. Intell..

[20]  Michael Luby,et al.  Approximating Probabilistic Inference in Bayesian Belief Networks is NP-Hard , 1993, Artif. Intell..

[21]  David J. Spiegelhalter,et al.  Local computations with probabilities on graphical structures and their application to expert systems , 1990 .

[22]  D. Fryback.  Bayes' theorem and conditional nonindependence of data in medical diagnosis. , 1978, Computers and biomedical research, an international journal.

[23]  Seymour Sherman,et al.  Non-mean-square error criteria , 1958, IRE Trans. Inf. Theory.

[24]  Gregory Provan,et al.  Model Selection for Diagnosis and Treatment Using Temporal Influence Diagrams , 1994 .

[25]  Robert P. Goldman,et al.  From knowledge bases to decision models , 1992, The Knowledge Engineering Review.

[26]  Ross D. Shachter.  Evaluating Influence Diagrams , 1986, Oper. Res..

[27]  J. Bernardo.  Expected Information as Expected Utility , 1979 .