摘要:

In this paper, the modular combination of artificial neural nets is considered. A modular approach to combining can be contrasted with an ensemble-based approach in that it implies individual modules, each responsible for some specialist aspect of a task, as opposed to each approximating the same function. It is possible to characterize modular systems in terms of (i) reasons for the task decomposition, (ii) the method for accomplishing the decomposition and (iii) the relationship between the modules. These characteristics are considered in brief outlines of the papers in the issue. Reasons for task decomposition include the exploitation of specialist capabilities of individual nets, performance improvement, and making the system easier to understand and modify. Task decomposition may be either automatic (based on the blind application of a data partitioning algorithm) or explicit (based on prior knowledge of the task or the specialist capabilities of the modules), and the relationship between the modules may be successive, cooperative or supervisory.

ISSN:0954-0091    

DOI:10.1080/095400997116702

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

In this paper, analysis of the information content of discretely firing neurons in unsupervised neural networks is presented, where information is measured according to the network's ability to reconstruct its input from its output with minimum mean square Euclidean error. It is shown how this type of network can self-organize into multiple winner-take-all subnetworks, each of which tackles only a low-dimensional subspace of the input vector. This is a rudimentary example of a neural network that effectively subdivides a task into manageable subtasks.

ISSN:0954-0091    

DOI:10.1080/095400997116711

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

This paper introduces bootstrap error estimation for automatic tuning of parameters in combined networks, applied as front-end preprocessors for a speech recognition system based on hidden Markov models. The method is evaluated on a large-vocabulary (10 000 words) continuous speech recognition task. Bootstrap estimates of minimum mean squared error allow selection of speaker normalization models improving recognition performance. The procedure allows a flexible strategy for dealing with inter-speaker variability without requiring an additional validation set. Recognition results are compared for linear, generalized radial basis functions and multi-layer perceptron network architectures.

ISSN:0954-0091    

DOI:10.1080/095400997116720

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

In this article, three different methods for hybridization and specialization of real-time recurrent learning (RTRL)-based neural networks (NNs) are presented. The first approach consists of combining recurrent networks with feedforward networks. The second approach continues with the combination of multiple recurrent NNs. The last approach introduces the combination of connectionist systems with instructionist artificial intelligence techniques. Two examples are added to demonstrate properties and advantages of these techniques. The first example is a process diagnosis task where a hybrid NN is connected to a knowledge-based system. The second example is a NN consisting of different recurrent modules that is used to handle missing sensor data in a process modelling task.

ISSN:0954-0091    

DOI:10.1080/095400997116739

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

This paper describes a medical application of modular neural networks (NNs) for temporal pattern recognition. In order to increase the reliability of prognostic indices for patients living with the acquired immunodeficiency syndrome (AIDS), survival prediction was performed in a system composed of modular NNS that classified cases according to death in a certain year of follow-up. The output of each NN module corresponded to the probability of survival in a given year. Inputs were the values of demographic, clinical and laboratory variables. The results of the modules were combined to produce survival curves for individuals. The NNs were trained by backpropagation and the results were evaluated in test sets of previously unseen cases. We showed that, for certain combinations of NN modules, the performance of the prognostic index, measured by the area under the receiver operating characteristic curve, was significantly improved (p 0.05). We also used calibration measurements to quantify the benefits of combining NN modules, and show why, when and how NNs should be combined for building prognostic models.

ISSN:0954-0091    

DOI:10.1080/095400997116748

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

This paper introduces HART-S, a new modular neural network (NN) that can incrementally learn stable hierarchical clusterings of arbitrary sequences of input patterns by self-organization. The network is a cascade of adaptive resonance theory (ART) modules, in which each module learns to cluster the differences between the input pattern and the selected category prototype at the previous module. Input patterns are first classified into a few broad categories, and successive ART modules find increasingly specific categories until a threshold is reached, the level of which can be controlled by a global parameter called 'resolution'. The network thus essentially implements a divisive (or splitting) hierarchical clustering algorithm: hence the name HART-S (for 'hierarchical ART with splitting'). HART-S is also compared and contrasted with HART-J (for 'hierarchical ART with joining'), another variant that was proposed earlier by the first author. The network dynamics are specified and some useful properties of both networks are given and then proven. Experiments were carried out on benchmark data sets to demonstrate the representational and learning capabilities of both networks and to compare the developed clusterings with those of two classical methods and a conceptual clustering algorithm. A brief survey of related NN models is also provided.

ISSN:0954-0091    

DOI:10.1080/095400997116757

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

We discuss a simple strategy aimed at improving neural network prediction accuracy, based on the combination of predictions at varying resolution levels of the domain under investigation (here: time series). First, a wavelet transform is used to decompose the time series into varying scales of temporal resolution. The latter provides a sensible decomposition of the data so that the underlying temporal structures of the original time series become more tractable. Then, a dynamical recurrent neural netork is trained on each resolution scale with the temporal-recurrent backpropagation algorithm. By virtue of its internal dynamic, this general class of dynamic connections network approximates the underlying law governing each resolution level by a system of non-linear difference equations. The individual wavelet scale forecasts are afterwards recombined to form the current estimate. The predictive ability of this strategy is assessed with the sunspot series.

ISSN:0954-0091    

DOI:10.1080/095400997116766

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor

摘要:

This paper proposes an application-independent method of automating learning rule parameter selection using a form of supervisor neural network (NN), known as a meta neural network (MNN), to alter the value of a learning rule parameter during training. The MNN is trained using data generated by observing the training of a NN and recording the effects of the selection of various parameter values. The MNN is then combined with a normal learning rule to augment its performance. Experiments are undertaken to see how this method performs by using it to adapt a global parameter of the resilient backpropagation and quickpropagation learning rules.

ISSN:0954-0091    

DOI:10.1080/095400997116775

出版商:Taylor & Francis Group    

年代:1997

数据来源: Taylor