We introduce a cost function for learning in feed-forward neural networks which is an explicit function of the internal representation inaddition 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 isan 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.

Interest in this algorithm waned in the 1970's after it was emphasized[Minsky andPapert, 1969] (1) that the class of problems solvable by a single half space was limited, and (2) that the Perceptron algorithm, although converging infinite time, did not converge in polynomial time. In the 1980's, however, it has become evident that there is no hope of providing a learning algorithm which can learn arbitrary functions in polynomial time and much research has thus been restricted to algorithms which learn a function drawn from a particular class of functions. Moreover, learning theory has focused on protocols like that of [Valiant, 1984] where we seek to classify, not a fixed set of examples, but examples drawn from a probability distribution. This allows a natural notion of "generalization" . There are very few classes which have yet been proven learnable in polynomial time, and one of these is the class of half spaces. Thus there is considerable theoretical interest now in studying the problem of learning a single half space, and so it is natural to reexamine the Percept ron algorithm within the formalism of Valiant. The Perceptron Algorithm Is Fast for Non-Malicious Distributions 677 In Valiant's protocol, a class of functions is called learnable if there is a learning algorithm which works in polynomial time independent of the distribution D generating the examples. Under this definition the Perceptron learning algorithm is not a polynomial time learning algorithm. However we will argue in section 2 that this definition is too restrictive.

A new learning algorithm, Learning by Choice of Internal Represetations (CHIR),was recently introduced. Whereas many algorithms reduce the learning process to minimizing a cost function over the weights, our method treats the internal representations as the fundamental entities to be determined. The algorithm applies a search procedure in the space of internal representations, and a cooperative adaptation of the weights (e.g. by using the perceptron learning rule). Since the introduction of its basic, single output version, theCHIR algorithm was generalized to train any feed forward network of binary neurons. Here we present the generalised version of the CHIR algorithm, and further demonstrate its versatility by describing how it can be modified in order to train networks with binary ( 1) weights. Preliminary tests of this binary version on the random teacher problem are also reported.

A large fraction of recent work in artificial neural nets uses multilayer perceptrons trained with the back-propagation algorithm described by Rumelhart et.

The concept of the stochastic Boltzmann machine (BM) is auractive for decision making and pattern classification purposes since the probability of attaining the network states is a function of the network energy. Hence, the probability of attaining particular energy minima may be associated with the probabilities of making certain decisions (or classifications). However, because of its stochastic nature, the complexity of the BM is fairly high and therefore such networks are not very likely to be used in practice. In this paper we suggest a way to alleviate this drawback by converting the stochastic BM into a deterministic network which we call the Boltzmann Perceptron Network (BPN). The BPN is functionally equivalent to the BM but has a feed-forward structure and low complexity.

A large fraction of recent work in artificial neural nets uses multilayer perceptrons trained with the back-propagation algorithm described by Rumelhart et.

Howard and M. A. Huckvale, "Acoustic-phonetic attribute determination using multi-layer perceptrons", IEEE Colloquium Digest 1988/11.

The use of a one-stage dynamic programming algorithm for connected word recognition.

This study evaluates the performance of the multilayer-perceptron and the frequency-sensitive competitive learning network in identifying five commercial aircraft from radar backscatter measurements. The performance of the neural network classifiers is compared with that of the nearest-neighbor and maximum-likelihood classifiers. Our results indicate that for this problem, the neural network classifiers are relatively insensitive to changes in the network topology, and to the noise level in the training data. While, for this problem, the traditional algorithms outperform these simple neural classifiers, we feel that neural networks show the potential for improved performance.

The application of neural networks to the demodulation of spread-spectrum signals in a multiple-access environment is considered. This study is motivated in large part by the fact that, in a multiuser system, the conventional (matched ter) fil receiver suffers severe performance degradation as the relative powers of the interfering signals become large (the "near-far" problem). Furthermore, the optimum receiver, which alleviates the near-far problem, is too complex to be of practical use. Receivers based on multi-layer perceptrons are considered as a simple and robust alternative to the optimum solution. The optimum receiver is used to benchmark the performance of the neural net receiver; in particular, it is proven to be instrumental in identifying the decision regions of the neural networks. The back-propagation algorithm and a modified version of it are used to train the neural net. An importance sampling technique is introduced to reduce the number of simulations necessary to evaluate the performance of neural nets.