Over the last five years or so, neural network (NN)-based approaches have been steadily gaining performance and popularity for a wide range of optical character recognition (OCR) problems, from isolated digit recognition to handprint recognition. We present an NN classification scheme based on an enhanced multilayer perceptron (MLP) and describe an end-to-end system for form-based handprint OCR applications designed by the National Institute of Standards and Technology (NIST) Visual Image Processing Group. The enhancements to the MLP are based on (i) neuron activations functions that reduce the occurrences of singular Jacobians; (ii) successive regularization to constrain the volume of the weight space; and (iii) Boltzmann pruning to constrain the dimension of the weight space. Performance characterization studies of NN systems evaluated at the first OCR systems conference and the NIST form-based handprint recognition system are also summarized.

Two-dimensional simulations of MOSFET's are widely used for the design of short-channel transistors used in VLSI circuits. These models use low order methods of discretization of solution variables. In this paper, a method of current calculation is presented which works with these methods and yields good accuracy. The method uses integration of the solution variables, rather than differentiation, and is similar to applying Ohm's law in two dimensions.

A successful portable version of a Fortran subprogram to find the Euclidean norm of an n-vector, I] x II = (~-~.~-1x,2) 11~, is described.Overflow and underflow are avoided.A program implementing the algorithm is included.Key Words and Phrases: error bounds, Euclidean norm, machine constants, overflow, portability, underflow CR Categories: 5.11, 5.14 Then [I x I] = xma~ a 1/~.This procedure requires two passes over the data vector, which is unnecessarily slow, especially for long vectors on paged machines.(Tested General