No abstract

The generalization ability of an MLP network has been shown to be related to both the number and magnitudes of the network weights. Thus, there exists a tension between employing networks with few weights that have relatively large magnitudes, and networks with a greater number of weights with relatively small magnitudes. The analysis presented in this paper indicates that large magnitudes for network weights potentially increase the propensity of a network to interpolate poorly.Experimental results indicate that when bounds are imposed on network weights, the backpropagation algorithm is capable of discovering networks with small weight magnitudes that retain their expressive power and exhibit good generalization.

Continuous records of annual landings and fishing effort exist in the Atlantic purse‐seine fishery for Atlantic menhaden Brevoortia tyrannus since 1940 and the Gulf of Mexico fishery for Gulf menhaden B. patronus since 1946. Currently, year‐ahead forecasts of landings from these species‐specific fisheries separated by the Florida peninsula are provided to the industry by means of multiple‐linear‐regression models that relate landings and effort over the data series. Here, we compare three methods for this purpose—multiple regression, time series, and artificial neural networks—to determine whether forecast accuracy can be increased. Best‐fit models were developed with each method for each fishery, and then 10‐year retrospective analyses of 1‐year‐ahead catch forecasts were compared among the three methods. In general, multiple‐regression and artificial neural network models were similar in their fit to the data series and both were better than time series models, judging from the Akaike information criterion, the correlation between observed and predicted catches, the mean prediction error, and the root mean square error of prediction. A 10‐year retrospective analysis (1993–2002) of 1‐year‐ahead catch forecasts indicates that the three methods provided similar within‐stock mean absolute forecast errors (19–21% in the Atlantic and 15–20% in the Gulf), with generally better forecasts for the Gulf fishery. Overall, multiple‐regression and artificial neural network models provide lower average catch forecast errors and better fits to the fishery data, whereas similar forecast errors are provided by a univariate time series model (autoregressive integrated moving average model) in the Atlantic and a multivariate time series model (state space model) in the Gulf.

Multilayer perceptron (MLP) networks trained using backpropagation can be slow to converge in many instances. The primary reason for slow learning is the global nature of backpropagation. Another reason is the fact that a neuron in an MLP network functions as a hyperplane separator and is therefore inefficient when applied to classification problems in which decision boundaries are nonlinear. This paper presents a data representational approach that addresses these problems while operating within the framework of the familiar backpropagation model. We examine the use of receptors with overlapping receptive fields as a preprocessing technique for encoding inputs to MLP networks. The proposed data representation scheme, termed ensemble encoding, is shown to promote local learning and to provide enhanced nonlinear separability. Simulation results for well known problems in classification and time-series prediction indicate that the use of ensemble encoding can significantly reduce the time required to train MLP networks. Since the choice of representation for input data is independent of the learning algorithm and the functional form employed in the MLP model, nonlinear preprocessing of network inputs may be an attractive alternative for many MLP network applications.