We present an application of back-propagation networks to handwritten digit recognition. Minimal preprocessing of the data was required, but architecture of the network was highly constrained and specifically designed for the task. The input of the network consists of normalized images of isolated digits. The method has 1 % error rate and about a 9% reject rate on zipcode digits provided by the U.S. Postal Service. 1 INTRODUCTION The main point of this paper is to show that large back-propagation (BP) networks can be applied to real image-recognition problems without a large, complex preprocessing stage requiring detailed engineering. Unlike most previous work on the subject (Denker et al., 1989), the learning network is directly fed with images, rather than feature vectors, thus demonstrating the ability of BP networks to deal with large amounts of low level information. Previous work performed on simple digit images (Le Cun, 1989) showed that the architecture of the network strongly influences the network's generalization ability. Good generalization can only be obtained by designing a network architecture that contains a certain amount of a priori knowledge about the problem. The basic design principle is to minimize the number of free parameters that must be determined by the learning algorithm, without overly reducing the computational power of the network.

Fernando J. Pineda Center for Microelectronics Technology Jet Propulsion Laboratory California Institute of Technology Pasadena, CA 91109 ABSTRACT A comparison of algorithms that minimize error functions to train the trajectories of recurrent networks, reveals how complexity is traded off for causality. These algorithms are also related to time-independent fonnalisms. It is suggested that causal and scalable algorithms are possible when the activation dynamics of adaptive neurons is fast compared to the behavior to be learned. Standard continuous-time recurrent backpropagation is used in an example. 1 INTRODUCTION Training the time dependent behavior of a neural network model involves the minimization of a function that measures the difference between an actual trajectory and a desired trajectory. The standard method of accomplishing this minimization is to calculate the gradient of an error function with respect to the weights of the system and then to use the gradient in a minimization algorithm (e.g.

The mapping of the back-propagation and mean field theory learning algorithms onto a generic 2-D SIMD computer is described. This architecture proves to be very adequate for these applications since efficiencies close to the optimum can be attained. Expressions to find the learning rates are given and then particularized to the DAP array procesor.

We have used information-theoretic ideas to derive a class of practical and nearly optimal schemes for adapting the size of a neural network. By removing unimportant weights from a network, several improvements can be expected: better generalization, fewer training examples required, and improved speed of learning and/or classification. The basic idea is to use second-derivative information to make a tradeoff between network complexity and training set error. Experiments confirm the usefulness of the methods on a real-world application. 1 INTRODUCTION Most successful applications of neural network learning to real-world problems have been achieved using highly structured networks of rather large size [for example (Waibel, 1989; Le Cun et al., 1990a)]. As applications become more complex, the networks will presumably become even larger and more structured.

A short account is given of various investigations of neural network properties, beginning with the classic work of McCulloch & Pitts. Early work on neurodynamics and statistical mechanics, analogies with magnetic materials, fault tolerance via parallel distributed processing, memory, learning, and pattern recognition, is described.

We study networks of spiking neurons in which spikes are fired as a Poisson process. The state of a cell is determined by the instantaneous firing rate, and in the limit of high firing rates our model reduces to that studied by Hopfield. We find that the inclusion of spiking results in several new features, such as a noise-induced asymmetry between "on" and "off" states of the cells and probability currents which destroy the usual description of network dynamics in terms of energy surfaces. Taking account of spikes also allows us to calibrate network parameters such as "synaptic weights" against experiments on real synapses. Realistic forms of the post synaptic response alters the network dynamics, which suggests a novel dynamical learning mechanism.

The paper suggests a statistical framework for the parameter estimation problem associated with unsupervised learning in a neural network, leading to an exploratory projection pursuit network that performs feature extraction, or dimensionality reduction.

A new concept for unsupervised learning based upon examples introduced to the neural network is proposed. Each example is considered as an interpolation node of the velocity field in the phase space. The velocities at these nodes are selected such that all the streamlines converge to an attracting set imbedded in the subspace occupied by the cluster of examples. The synaptic interconnections are found from learning procedure providing selected field. The theory is illustrated by examples. This paper is devoted to development of a new concept for unsupervised learning based upon examples introduced to an artificial neural network.

ABSTRACT The CHAC storage scheme has been used as a basis for a software implementation of an associative .emory A major disadvantage of this CHAC-concept is that the degree of local generalization (area of interpolation) is fixed. This paper deals with an algorithm for self-organizing variable generalization for the AKS, based on ideas of T. Kohonen. 1 INTRODUCTION For several years research at the Department of Control Theory and Robotics at the Technical University of Darmstadt has been concerned with the design of a learning real-time control loop with neuron-like associative memories (LERNAS) A Self-organizing Associative Memory System for Control Applications 333 for the control of unknown, nonlinear processes (Ersue, Tolle, 1988). This control concept uses an associative memory system AHS, based on the cerebellar cortex model CHAC by Albus (Albus, 1972), for the storage of a predictive nonlinear process model and an appropriate nonlinear control strategy (Figure 1). Figure 1: The learning control loop LERNAS One problem for adjusting the control loop to a process is, however, to find a suitable set of parameters for the associative memory. The parameters in question determine the degree of generalization within the memory and therefore have a direct influence on the number of training steps required to learn the process behaviour. For a good performance of the control loop it· is desirable to have a very small generalization around a given setpoint but to have a large generalization elsewhere.

The brain represents the skin surface as a topographic map in the somatosensory cortex. This map has been shown experimentally to be modifiable in a use-dependent fashion throughout life. We present a neural network simulation of the competitive dynamics underlying this cortical plasticity by detailed analysis of receptive field properties of model neurons during simulations of skin coactivation, cortical lesion, digit amputation and nerve section. 1 INTRODUCTION Plasticity of adult somatosensory cortical maps has been demonstrated experimentally in a variety of maps and species (Kass, et al., 1983; Wall, 1988). This report focuses on modelling primary somatosensory cortical plasticity in the adult monkey. We model the long-term consequences of four specific experiments, taken in pairs. With the first pair, behaviorally controlled stimulation of restricted skin surfaces (Jenkins, et al., 1990) and induced cortical lesions (Jenkins and Merzenich, 1987), we demonstrate that Hebbian-type dynamics is sufficient to account for the inverse relationship between cortical magnification (area of cortical map representing a unit area of skin) and receptive field size (skin surface which when stimulated excites a cortical unit) (Sur, et al., 1980; Grajski and Merzenich, 1990). These results are obtained with several variations of the basic model. We conclude that relying solely on cortical magnification and receptive field size will not disambiguate the contributions of each of the myriad circuits known to occur in the brain. With the second pair, digit amputation (Merzenich, et al., 1984) and peripheral nerve cut (without regeneration) (Merzenich, ct al., 1983), we explore the role of local excitatory connections in the model Neural Network Simulation of Somatosensory Representational Plasticity S3