A neural network approach to statistical pattern classification by 'semiparametric' estimation of probability density functions

摘要: A method for designing near-optimal nonlinear classifiers, based on a self-organizing technique estimating probability density functions when only weak assumptions are made about the densities, is described. The avoids disadvantages of other existing methods by parametrizing set component densities from which actual constructed. parameters optimized algorithm, reducing to minimum labeling design samples. All required computations realized with simple sum-of-product units commonly used in connectionist models. approximations produced illustrated two dimensions multispectral image classification task. practical use small speech recognition problem. Related issues invariant projections, cross-class pooling data, and subspace partitioning discussed. >