Abstract
In the construction of a probabilistic network from observed data, the fundamental assumption that the variables starting from the same parent are conditionally independent can be met by introduction of hidden node. The conditional probability matrices for the hidden node, linking observed nodes, can be determined by the gradient descent method. In our previous work in the field of medical diagnosis based on computer vision, we demonstrated that this technique does yield improved results, however the success depends on the choice of initial values for the conditional probabilities. In this paper we introduce methods that provide a good initial estimation of the hidden variables and improve our learning algorithm. With these improvements, we have found that the first solution obtained by the training algorithm is a near optimal solution, and we have eliminated the need for multiple restarts.
Preview
Unable to display preview. Download preview PDF.
References
C K Chow, C N Liu, “Approximating Discrete Probability Distributions with Dependence Trees.” IEEE Transactions on Information Theory, Vol. 14, No. 3, pp. 462–467
D. A. Clark, “Numerical and symbolic approaches to uncertainty management in AI.” Artificial Intelligence Review 4 (1990), pp.109–146
J. B. Hampshire, B. Pearlmutter, “Equivalence proofs for multi-layer perceptron classifiers and the Bayesian discriminant function.” Connectonist Models, Proceedings of the 1990 summer school, edited by D. S. Touretzky, J. L. Elman, T J. Sejnowski, G. E. Hinton
G. N. Khan, “Machine vision for endoscope control and navigation.” Ph.D. Thesis, Imperial College, London.
Kwoh, Chee Keong and Gillies, Duncan Fyfe, “Using Hidden Nodes in Probabilistic Networks.” Submitted paper.
P. J. M. van Larrhoven and E. H. L. Aarts, “Simulated Annealing: Theory and Applications.” D Reidel Publishing Company, 1987.
S. P. Luttrell, “Adaptive Bayesian networks.” SPIE Vol. 1706, pp. 140–151, 1992
Judea Pearl. “Probabilistic reasoning in intelligent systems: networks of plausible inference.” Morgan Kaufmann, c1988
J. Pearl, T S Verma, “A Theory of Inferred Causation.” 2nd Conference on the Principles of Knowledge Representation and Reasoning. pp441–452.
L. E. Sucar, D. F. Gillies, D. A. Gillies, “Objective probabilities in expert systems.” Artificial Intelligence 61 (1993), pp.187–208
Lei Xu, Judea Pearl, “Structuring Causal Tree Models with Continuous Variables.” Uncertainty in AI 3, edited by L. N. Kanai, T. S. Levitt and J. F. Lemmer. pp 209–219
B. Widrow, S. D. Stearns, “Adaptive Signal Processing.” Prentice Hall, cl985
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1995 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Kwoh, C.K., Gillies, D.F. (1995). Estimating the initial values of unobservable variables in visual probabilistic networks. In: Hlaváč, V., Šára, R. (eds) Computer Analysis of Images and Patterns. CAIP 1995. Lecture Notes in Computer Science, vol 970. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-60268-2_313
Download citation
DOI: https://doi.org/10.1007/3-540-60268-2_313
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-60268-2
Online ISBN: 978-3-540-44781-8
eBook Packages: Springer Book Archive