By Topic

# 1989 First IEE International Conference on Artificial Neural Networks, (Conf. Publ. No. 313)

## Filter Results

Displaying Results 1 - 25 of 87
• ### Estimating hidden units for two-layer perceptrons

Publication Year: 1989, Page(s):120 - 124
Cited by:  Papers (25)
| | PDF (241 KB)

A method of estimating the number of hidden units required by a two-layer perceptron learning binary mappings using back propagation of error signals is presented. In order to obtain an estimate of the number of hidden units for a fully connected net with n output units, it is necessary to obtain an estimate of the number of 'conflicts' contained in the individual binary responses that must be lea... View full abstract»

• ### Diffusion learning for the multilayer perceptron

Publication Year: 1989, Page(s):390 - 394
Cited by:  Papers (5)  |  Patents (1)
| | PDF (191 KB)

A method of nonconvex optimisation based on simulating diffusion processes is discussed and applied to learning in the multilayer perceptron. It is compared with simulated annealing and back propagation of error as an optimisation learning tool. Results indicate significant improvement on these traditional learning methods. The method may be easily implemented in existing hardware.<> View full abstract»

• ### A neural network approach to the computation of visual algorithms

Publication Year: 1989, Page(s):67 - 70
| | PDF (188 KB)

This paper presents a method that combines a parallel iterative network for the computation of an early vision algorithm with a connectionist network which is used for the determination of discontinuities in images. Both networks are implemented on a parallel SIMD processor array, the ICL Distributed Array processor View full abstract»

• ### A general purpose digital architecture for neural network simulations

Publication Year: 1989, Page(s):62 - 66
Cited by:  Papers (1)  |  Patents (16)
| | PDF (564 KB)

A general-purpose neuro-chip for the resolution and learning stages of neural computing is described. Parallelism on input neurons is achieved for output neuron states evaluation and for synaptic weights updating. The latter follows a Hebb-like local formula. The digital realization guarantees exact integer calculations. According to simulations and theoretical considerations, 16-bit precision on ... View full abstract»

• ### Self-organization based on the second maximum entropy principle

Publication Year: 1989, Page(s):12 - 16
Cited by:  Papers (3)
| | PDF (192 KB)

The article formulates an optimal mapping from a continuous onto a discrete random variable by introducing the second maximum entropy principle, complementary to the Gibbsian one. The mapping corresponds to the self-organization of a community of formal neurons. The derived properties of the interaction between neurons are similar to those in biological neural networks View full abstract»

• ### The implementation of hardware neural net systems

Publication Year: 1989, Page(s):57 - 61
Cited by:  Papers (3)
| | PDF (268 KB)

Describes a multilayer pipelined digital architecture suitable for the implementation of large neural nets (LN) for vision applications. It can also be used to do some pre-filtering, such as pixel averaging, by setting weight values appropriately. A 1024 node processor with a clock rate of 10 MHz can operate on an input vector consisting of 32×32 8 bit pixels in 102 μs. It can therefore p... View full abstract»

• ### Decision feedback equalization using neural network structures

Publication Year: 1989, Page(s):125 - 128
Cited by:  Papers (6)  |  Patents (2)
| | PDF (184 KB)

Describes a new approach for a decision feedback equalizer (DFE) using the multi-layer perceptron structure for equalization in digital communications systems. Results indicate that the multi-layer perceptron DFE provides better BER (bit error rate) performance relative to the standard least mean square DFE structure, especially in high noise conditions View full abstract»

• ### A comparative study of the Kohonen and multiedit neural net learning algorithms

Publication Year: 1989, Page(s):7 - 11
| | PDF (280 KB)

This paper presents a comparative evaluation of the multiedit/condensing and Kohonen neural net learning algorithms using a speaker-independent speech recognition problem as a test vehicle. Both approaches attempt to cover the subspaces associated with respective pattern classes by a small number of reference vectors for subsequent nearest neighbour classification of unknown patterns. Several impo... View full abstract»

• ### Can perceptrons find Lyapunov functions?-an algorithmic approach to systems stability

Publication Year: 1989, Page(s):364 - 368
| | PDF (188 KB)

A machine implementable algorithm is developed for determining Lyapunov functions. The method is based on the perceptron algorithm and so is easily implemented on a neural network View full abstract»

• ### An implementation of fully analogue sum-of-product neural models in VLSI

Publication Year: 1989, Page(s):52 - 56
Cited by:  Papers (6)
| | PDF (232 KB)

Neural networks that use digital or partly digital processing units have a restricted set of applications, and also are constrained by the set of learning rules that can be used with them. Analogue networks have a greater flexibility in their learning algorithms and a larger domain of problems which they can solve. The flexibility of an analogue output capability can also be used to give an idea o... View full abstract»

• ### Infrared search and track signal processing: a potential application of artificial neural computing

Publication Year: 1989, Page(s):270 - 274
| | PDF (236 KB)

A requirement of modern airborne infrared search and track (IRST) systems is to detect targets at long distances, often in the presence of heavy cloud clutter in sky scenes, or possibly ground clutter when the IR sensor is looking down. A number of spatial processing algorithms have been tested in this study, including (1) an LMS filter and (2) an artificial neural network trained to recognize poi... View full abstract»

• ### Neural networks and Hough transform for pattern recognition

Publication Year: 1989, Page(s):81 - 85
| | PDF (152 KB)

The Hough transform (HT) is a very efficient technique in pattern recognition for performing template matching of curves and is very suitable for implementation in parallel hardware. The paper presents the HT as a neural network (NN). This fact suggests that the HT can be used to determine the parameters of the neurons' responsible for the detection of patterns to be recognized. After a brief rev... View full abstract»

• ### Hierarchical self-organising networks

Publication Year: 1989, Page(s):2 - 6
Cited by:  Papers (5)
| | PDF (344 KB)

A training scheme for a nonparametric neural network, which leads to the vector quantiser is derived. Then, the robust hidden layer principle is introduced in order to relate the vector quantiser to self-organising neural networks. Finally, it is demonstrated how hierarchical self-organising neural networks may be constructed by further application of the robust hidden layer principle View full abstract»

• ### Training networks with discontinuous activation functions

Publication Year: 1989, Page(s):361 - 363
| | PDF (168 KB)

This paper presents a learning algorithm which may be used to train networks whose neurons may have discontinuous or nondifferentiable activation functions. The algorithm has been demonstrated using several different neuron activation functions. Although it shares several features with the error back-propagation algorithm, the heuristic derivation presented does not appeal to the highly mathematic... View full abstract»

• ### A/D conversion and analog vector quantization using neural network models

Publication Year: 1989, Page(s):324 - 328
| | PDF (312 KB)

A neural network model for an A/D converter shown in the paper is a good example where the computation complexity is reduced when some of the inter-connections and coefficients are removed. The strategy is to use a hierarchical structure which leads to a multi-layer feedforward realization. The analog pipeline connections used in the artificial neural net model for A/D conversion will not be slowe... View full abstract»

• ### First IEE International Conference on Artificial Neural Networks (Conf. Publ. No.313)

Publication Year: 1989
| | PDF (24 KB)

The following topics were dealt with: internal representations; self-organising and feedback networks; implementations; applications to vision; speech processing, signal and data processing, image processing, A/D conversion, and Lyapunov function finding; pattern recognition; architectures; multilayer perceptrons; AI; learning systems; network evolution; event identification; searching; graph matc... View full abstract»

• ### Ferroelectric connections for IC neural networks

Publication Year: 1989, Page(s):47 - 51
Cited by:  Papers (3)
| | PDF (240 KB)

The application of ferroelectric thin-film technology to the construction of electrically-alterable synapse elements for artificial neural net applications is discussed. The ferroelectric film resides above the active circuitry, allowing high density. Firstly, a binary connection scheme supporting limited connectivity but utilizing a ferroelectric characteristic which is currently used in producti... View full abstract»

• ### A model of early vertebrate visual processing

Publication Year: 1989, Page(s):115 - 119
| | PDF (332 KB)

A mathematical analysis is given of an analog model of retinal processing constructed recently in terms of a resistive lattice network by Mead and Mahowald (1988). The basic equations are written down for a general lattice, and their continuum limit described. For linear resistors the general solution is given in terms of an arbitrary varying illumination input; some special cases are discussed in... View full abstract»

• ### A new learning paradigm for neural networks

Publication Year: 1989, Page(s):346 - 350
Cited by:  Papers (1)
| | PDF (280 KB)

Introduces a new way of inferring the structure of a temporal neural network from a set of training data. The approach is to learn a grammar which describes and generalises the input patterns, and then to map this grammar onto a connectionist architecture so allowing the network topology to be specialised to the training data. The resulting network has as many levels as are necessary, and arbitrar... View full abstract»

• ### A parallel architecture for nonlinear adaptive filtering and pattern recognition

Publication Year: 1989, Page(s):265 - 269
Cited by:  Patents (1)
| | PDF (292 KB)

In the late 1950s there were two particular, closely related developments in information processing: the perceptron, based on the work of Rosenblatt (1961) and the adaptive noise canceller of Widrow and Hoff (1960). Subsequently the two ideas progressed in different ways under the patronage of different sections of the scientific community. The present paper centres on an assertion of their origin... View full abstract»

• ### Optical character recognition using artificial neural networks

Publication Year: 1989, Page(s):191 - 195
Cited by:  Papers (1)
| | PDF (304 KB)

Optical character recognition is examined to find a general framework by which it can be realized. A hierarchical cone' with feature extraction layers of increasing sophistication is described. The system, unlike the artificial neural net examples in the literature, does not use one network only. Allowing recognition to take place in parallel over different representations of the same symbol intr... View full abstract»

• ### Evolution equations for neural networks with arbitrary spatial structure

Publication Year: 1989, Page(s):238 - 241
Cited by:  Patents (1)
| | PDF (176 KB)

The question of how to describe networks in which the range of connections is restricted or in which the connection density is not uniform in space is still unanswered. The purpose of the paper is to remedy this situation for the case where the range of the connections is large compared to the average distance between neighbouring neurons. The number of connections to and from each neuron are assu... View full abstract»

• ### Optical flow estimation by using the artificial neural network under multi-layers

Publication Year: 1989, Page(s):76 - 80
Cited by:  Patents (1)
| | PDF (332 KB)

A Hopfield model for computing optical flow is presented. A set of features describing the local intensity structure along the principal directions is used to measure the matching between the two local neighborhoods in the successive frames. The energy function can be derived based on the match measure and regularized by adding the Tikhonov stabilizer of the smoothness constraints. This energy fun... View full abstract»

• ### On the significance of internal representations in neural networks

Publication Year: 1989
| | PDF (32 KB)

Summary form only given. For a straightforward materialization of internal representations, semantic networks are suggested. In their original form they comprise a graph structure with nodes and links. The nodes may stand for items or concepts (sets of attributes), whereas the links usually indicate relations. In view of the contemporary neurophysiological data, such a degree of specificity and sp... View full abstract»

• ### Image processing with optimum neural networks

Publication Year: 1989, Page(s):374 - 377
| | PDF (384 KB)

Although neurons in intermediate layers (hidden units) are very important components of most neural networks, there is a lack of general rules specifying how many hidden layers and how many hidden units per layer should be used to achieve optimum performance of a network. This lack of rules has its roots in the difficulty in judging the performance of the hidden units. The most widely used trainin... View full abstract»