Presents the results found when exploring the ability of a particular neural network to model and predict electrocardiograms, a kind of signal believed to be mathematically chaotic. Two concepts are embedded in the design of the presented model: Lyapunov exponents and harmonic generators. The term "harmonic generator" is used to describe a 3-node, fully connected recurrent neural network trained to produce a sine wave with a specific frequency and amplitude. Harmonic generators are able to reproduce accurately sine trajectories for long periods of time without using any external inputs, Our network called the hybrid-complex neural network was able to represent some of the dynamics of the system, showing fairly good short-term prediction and some oscillation during the long-term prediction, even when the external inputs came from previous predictions of the network. These characteristics are not observed in plain feed-forward or recurrent neural networks.

In this study, we have used the principle of competitive learning to develop an iterative algorithm for image recovery and segmentation. Within the framework of Markov random fields (MRFs), the image recovery problem is transformed to the problem of minimization of an energy function; A local update rule for each pixel point is then developed in a stepwise fashion and is shown to be a gradient descent rule for an associated global energy function. The relationship of the update rule to Kohonen's update rule is shown. Quantitative measures of edge preservation and edge enhancement for synthetic images are introduced. As compared to recently published results using mean field approximation, our algorithm shows consistently better performance in edge preservation and comparable performance in enhancing within the boundaries. These results are based on simulation experiments on a set of synthetic images corrupted by Gaussian noise and on a set of real images.

Dynamic task allocation for distributed computing systems (DCS) is an important goal to be achieved for engineering applications. The purpose of dynamic task allocation is to increase the system throughput in a dynamic environment, which can be done by balancing the utilization of computing resources and minimizing communication between processors during run time. In this paper, we propose two dynamic task allocation models which are: 1) the clustering simulated annealing model (CSAM); and 2) the mean field annealing model (MFAM). Both of these models combine characteristics of statistical and deterministic approaches. These models provide the rapid convergence characteristic of the deterministic approaches while preserving the solution quality afforded by simulated annealing. Simulation results of the CSAM and MFAM provide a stable and balanced system with 50% and 10% of the convergence time needed by simulated annealing, respectively. The results of this research are important in that it presents the feasibility of applying statistically based task allocation models on large DCSs in a dynamic environment. Solutions of these models depend on the annealing process instead of the structures of the input data, providing the possibility of obtaining better solutions by using more efficient computing hardware.

Concerns the problem of finding weights for feed-forward networks in which threshold functions replace the more common logistic node output function. The advantage of such weights is that the complexity of the hardware implementation of such networks is greatly reduced. If the task to be learned does not change over time, it may be sufficient to find the correct weights for a threshold function network off-line and to transfer these weights to the hardware implementation. This paper provides a mathematical foundation for training a network with standard logistic function nodes and gradually altering the function to allow a mapping to a threshold unit network. The procedure is analogous to taking the limit of the logistic function as the gain parameter goes to infinity. It is demonstrated that, if the error in a trained network is small, a small change in the gain parameter will cause a small change in the network error. The result is that a network that must be implemented with threshold functions can first be trained using a traditional back propagation network using gradient descent, and further trained with progressively steeper logistic functions. In theory, this process could require many repetitions. In simulations, however, the weights have be successfully mapped to a true threshold network after a modest number of slope changes. It is important to emphasize that this method is only applicable to situations for which off-line learning is appropriate.