论文信息 - Pattern classification by stochastic neural network with missing data

Pattern classification by stochastic neural network with missing data

In this paper, pattern classification by stochastic neural networks is considered. This model is also called a Gaussian mixture model. When missing data exist in the training data, it is usual to remove incomplete instants. Here we take another approach, where the missing elements are estimated by using the conditional expectation based on the estimated model by using the EM algorithm. It is shown by using Fisher's Iris data that this approach is superior to removing incomplete data.

[1] Donald F. Specht,et al. Probabilistic neural networks and general regression neural networks , 1996 .

[2] Roy L. Streit,et al. Maximum likelihood training of probabilistic neural networks , 1994, IEEE Trans. Neural Networks.

[3] John Moody,et al. Fast Learning in Networks of Locally-Tuned Processing Units , 1989, Neural Computation.

[4] Michael I. Jordan,et al. On Convergence Properties of the EM Algorithm for Gaussian Mixtures , 1996, Neural Computation.

[5] D. Rubin,et al. Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper , 1977 .

[6] Geoffrey E. Hinton,et al. Learning internal representations by error propagation , 1986 .

[7] E. Parzen. On Estimation of a Probability Density Function and Mode , 1962 .

[8] R. Fisher. THE USE OF MULTIPLE MEASUREMENTS IN TAXONOMIC PROBLEMS , 1936 .

[9] Catherine Blake,et al. UCI Repository of machine learning databases , 1998 .