Casasent network density estimation

Density estimation is an integral component of many signal processing tasks such as pattern recognition, detection, and cluster analysis. The kernel density estimator's utility is limited in high dimensional spaces by the curse of dimensionality. Modha (1994) proposed a neural network density estimator that avoids this 'curse', but can potentially require many hidden units (and therefore a lot of computation) for even simple densities such as Gaussians. We propose an extension of Modha's neural net using higher order hidden units that allow simple densities to be modeled economically (as kernel estimators do), while retaining the flexibility to avoid the curse of dimensionality (as Modha's network does).

[1]  Halbert White,et al.  Parametric Statistical Estimation with Artificial Neural Networks: A Condensed Discussion , 1994 .

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

[3]  Donald F. Specht,et al.  Probabilistic neural networks , 1990, Neural Networks.

[4]  Andrew R. Barron,et al.  Universal approximation bounds for superpositions of a sigmoidal function , 1993, IEEE Trans. Inf. Theory.

[5]  G. D. Wilensky,et al.  The projection neural network , 1992, [Proceedings 1992] IJCNN International Joint Conference on Neural Networks.

[6]  Yeshaiahu Fainman,et al.  A learning law for density estimation , 1994, IEEE Trans. Neural Networks.

[7]  A. Sarajedini,et al.  The best of both worlds: Casasent networks integrate multilayer perceptrons and radial basis functions , 1992, [Proceedings 1992] IJCNN International Joint Conference on Neural Networks.

[8]  D. F. Specht,et al.  Enhancements to probabilistic neural networks , 1992, [Proceedings 1992] IJCNN International Joint Conference on Neural Networks.