We introduce a cost function for learning in feed-forward neural networks which is an explicit function of the internal representation in addition to the weights. The learning problem can then be formulated as two simple perceptrons and a search for internal representations. Back-propagation is recovered as a limit. The frequency of successful solutions is better for this algorithm than for back-propagation when weights and hidden units are updated on the same timescale i.e. once every learning step. 1 INTRODUCTION In their review of back-propagation in layered networks, Rumelhart et al. (1986) describe the learning process in terms of finding good "internal representations" of the input patterns on the hidden units. However, the search for these representations is an indirect one, since the variables which are adjusted in its course are the connection weights, not the activations of the hidden units themselves when specific input patterns are fed into the input layer. Rather, the internal representations are represented implicitly in the connection weight values. More recently, Grossman et al. (1988 and 1989)1 suggested a way in which the search for internal representations could be made much more explicit.

J.B. Pollack, Implications of Recursive Distributed Representations, Advances in Neural Information Systems 1, D.S. Touretzky (ed), Morgan Kaufmann, San Mateo, Ca, p. 527 (1989).

We show that localized activity patterns in a layer of cells, collective excitations, can induce the formation of modular structures in the anatomical connections via a Hebbian learning mechanism. The networks are spatially homogeneous before learning, but the spontaneous emergence of localized collective excitations and subsequently modularity in the connection patterns breaks translational symmetry. This spontaneous symmetry breaking phenomenon is similar to those which drive pattern formation in reaction-diffusion systems. We have identified requirements on the patterns of lateral connections and on the gains of internal units which are essential for the development of modularity. These essential requirements will most likely remain operative when more complicated (and biologically realistic) models are considered.

Data is usually available in the form of a set S consisting of {x,y} pairs, where x is a input vector and y is the corresponding output vector.

Eight neural net and conventional pattern classifiers (Bayesianunimodal Gaussian, k-nearest neighbor, standard back-propagation, adaptive-stepsize back-propagation, hypersphere, feature-map, learning vector quantizer, and binary decision tree) were implemented on a serial computer and compared using two speech recognition and two artificial tasks. Error rates were statistically equivalent on almost all tasks, but classifiers differed by orders of magnitude in memory requirements, training time, classification time, and ease of adaptivity. Nearest-neighbor classifiers trained rapidly but required the most memory. Tree classifiers provided rapid classification but were complex to adapt. Back-propagation classifiers typically required long training times and had intermediate memory requirements. These results suggest that classifier selection should often depend more heavily on practical considerations concerning memory and computation resources, and restrictions on training and classification times than on error rate.

In our effort to develop a modular neural system for invariant learning and recognition of 3D objects, we introduce here a new module architecture called an aspect network constructed around adaptive axo-axo-dendritic synapses. This builds upon our existing system (Seibert & Waxman, 1989) which processes 20 shapes and classifies t.hem into view categories (i.e., aspects) invariant to illumination, position, orientat.ion,

We have constructed a two axis camera positioning system which is roughly analogous to a single human eye. This Artificial-Eye (Aeye) combines the signals generated by two rate gyroscopes with motion information extracted from visual analysis to stabilize its camera. This stabilization process is similar to the vestibulo-ocular response (VOR); like the VOR, A-eye learns a system model that can be incrementally modified to adapt to changes in its structure, performance and environment. A-eye is an example of a robust sensory system that performs computations that can be of significant use to the designers of mobile robots. 1 Introduction We have constructed an "artificial eye" (A-eye), an autonomous robot that incorporates a two axis camera positioning system (figure 1). Like a the human oculomotor system, A-eye can estimate the rotation rate of its body with a gyroscope and estimate the rotation rate of its "eye" by measuring image slip

Recurrent nets are more powerful than feedforward nets because they allow simulation of dynamical systems. Everything from sine wave generators through computers to the brain are potential candidates, but to use recurrent nets to emulate dynamical systems we need learning algorithms to program them. Here I describe a new twist on an old algorithm for recurrent nets and compare it to its predecessors.

Given a set of input-output training samples, we describe a procedure for determining the time sequence of weights for a dynamic neural network to model an arbitrary input-output process. We formulate the input-output mapping problem as an optimal control problem, defining a performance index to be minimized as a function of time-varying weights. We solve the resulting nonlinear two-point-boundary-value problem, and this yields the training rule. For the performance index chosen, this rule turns out to be a continuous time generalization of the outer product rule earlier suggested heuristically by Hopfield for designing associative memories. Learning curves for the new technique are presented.

In this paper we describe the VLSI design and testing of a high capacity associative memory which we call the exponential correlation associative memory (ECAM). The prototype 3J.'-CMOS programmable chip is capable of storing 32 memory patterns of 24 bits each. The high capacity of the ECAM is partly due to the use of special exponentiation neurons, which are implemented via sub-threshold MOS transistors in this design. The prototype chip is capable of performing one associative recall in 3 J.'S.