Dynamic Competitive Probabilistic Principal Components Analysis

Abstract

We present a new neural model which extends the classical competitive learning (CL) by performing a Probabilistic Principal Components Analysis (PPCA) at each neuron. The model also has the ability to learn the number of basis vectors required to represent the principal directions of each cluster, so it overcomes a drawback of most local PCA models, where the dimensionality of a cluster must be fixed a priori. Experimental results are presented to show the performance of the network with multispectral image data.

Fractality analysis of frontal brain in major depressive disorder

“…The Probabilistic Neural Network (PPN) has been used in many applications as a simple and efficient classifier (Lopez-Rubio and Ortiz-de-Lazcano-Lobato, 2009;Adeli and Panakkat, 2009;. PNN assigns the test data to the class with maximum likelihood compared with other classes.…”

Head-Centric Disparity and Epipolar Geometry Estimation From a Population of Binocular Energy Neurons

“…Using PCA, we reduce the dimensionality to 500, which for this problem typically retains around 95% of the variance. PCA has been successfully applied in a variety of domains [20][21][22] and a large number of biologically plausible learning rules can be used to perform it. 23,24 For simplicity, and due to the high dimensionality of the population response, we use the memory-efficient iterative implicitly restarted Lanczos method 25 that is provided by Matlab's eigs function.…”

Image Compression Based on Generalized Principal Components Analysis and Simulated Annealing

“…In PCA approach, a lot of information in a dataset is placed into a reduced dimension data structure by projecting the entire dataset onto a sub-space generated by an orthonormal axes system (Ye & Li, 2004). The optimal axes system can be evaluated using Singular Values Decomposition (SVD) (Golub & Van Loan, 1996;Wu, Warwick, Ma, Gasson, Burgess, Pan, & Aziz, 2010;López-Rubio & Lazcano-Lobato, 2009;Tipping & Bishop, 1999a, 1999b. The reduced dimensions data structure are chosen so that important features of the data are captured with low-loss of information (Ye & Li, 2004).…”