Application of the recurrent multilayer perceptron in modelingcomplex process dynamics

A nonlinear dynamic model is developed for a process system, namely a heat exchanger, using the recurrent multilayer perceptron network as the underlying model structure. The perceptron is a dynamic neural network, which appears effective in the input-output modeling of complex process systems. Dynamic gradient descent learning is used to train the recurrent multilayer perceptron, resulting in an order of magnitude improvement in convergence speed over a static learning algorithm used to train the same network. In developing the empirical process model the effects of actuator, process, and sensor noise on the training and testing sets are investigated. Learning and prediction both appear very effective, despite the presence of training and testing set noise, respectively. The recurrent multilayer perceptron appears to learn the deterministic part of a stochastic training set, and it predicts approximately a moving average response of various testing sets. Extensive model validation studies with signals that are encountered in the operation of the process system modeled, that is steps and ramps, indicate that the empirical model can substantially generalize operational transients, including accurate prediction of instabilities not in the training set. However, the accuracy of the model beyond these operational transients has not been investigated. Furthermore, online learning is necessary during some transients and for tracking slowly varying process dynamics. Neural networks based empirical models in some cases appear to provide a serious alternative to first principles models.     

Education-Stratified Base-Rate Information on Discrepancy Scores Within and Between the Wechsler Adult Intelligence Scale-Third Edition and the Wechsler Memory Scale-Third Edition.

The Wechsler Adult Intelligence Scale--Third Edition (WAIS-III; D. Wechsler, 1997a) and the Wechsler Memory Scale--Third Edition (WMS-III; D. Wechsler, 1997b) are 2 of the most frequently used measures in psychology and neuropsychology. To facilitate the diagnostic use of these measures in the clinical decision-making process, this article provides information on education-stratified, directional prevalence rates (i.e., base rates) of discrepancy scores between the major index scores for the WAIS-III, the WMS-III, and between the WAIS-III and WMS-III. To illustrate how such base-rate data can be clinically used, this article reviews the relative risk (i.e., odds ratio) of empirically defined "rare" cognitive deficits in 2 of the clinical samples presented in the WAIS-III--WMS-III Technical Manual (The Psychological Corporation, 1997). (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The early restart algorithm

Consider an algorithm whose time to convergence is unknown (because of some random element in the algorithm, such as a random initial weight choice for neural network training). Consider the following strategy. Run the algorithm for a specific time T. If it has not converged by time T, cut the run short and rerun it from the start (repeat the same strategy for every run). This so-called restart mechanism has been proposed by Fahlman (1988) in the context of backpropagation training. It is advantageous in problems that are prone to local minima or when there is a large variability in convergence time from run to run, and may lead to a speed-up in such cases. In this article, we analyze theoretically the restart mechanism, and obtain conditions on the probability density of the convergence time for which restart will improve the expected convergence time. We also derive the optimal restart time. We apply the derived formulas to several cases, including steepest-descent algorithms.     

Recognition of multiunit neural signals

An essential step in studying nerve cell interaction during information processing is the extracellular microelectrode recording of the electrical activity of groups of adjacent cells. The recording usually contains the superposition of the spike trains produced by a number of neurons in the vicinity of the electrode. It is therefore necessary to correctly classify the signals generated by these different neurons. This paper considers this problem, and a new classification scheme is developed, which does not require human supervision. A learning stage is first applied on the beginning portion of the recording to estimate the typical spike shapes of the different neurons. As for the classification stage, a method is developed, which specifically considers the case when spikes overlap temporally. The method minimizes the probability of error, taking into account the statistical properties of the discharges of the neurons. The method is tested on a real recording as well as on synthetic data.     

A new generalized inverter based on the active hybrid

The building block for the new generalized immittance inverter is a 3-port known as the active hybrid network. One of its ports is bidirectional, and matches a design admittance y₀. The second port operates in the send, while the third port operates in the receive direction. The hybrid is constructed from an op-amp and a 5-port network composed of a few resistors and the design admittance y₀. There are two versions of the hybrid: the voltage-inverting (i.h.) and the noninverting hybrid (n.i.h.). The inverter consists of an i.h. and an n.i.h. connected at their unidirectional ports in a loop. The gyration admittances are actual driving-point admittances in the circuit, namely the design admittances y₀, y'₀ of the two hybrids. Such an inverter, when terminated at its output port by a load admittance y, will give at its input port an input admittance y_in = y₀y'₀/y. Expressions are derived for the y parameters of the inverter at higher frequencies, thus accounting for the finite bandwidth of the op-amp. Two special cases are particularly treated, and experimental results are obtained for them. The first is the case of the resistive gyrator, for which y₀ and y'₀ are pure conductances. The second is the case of the capacitive gyrator, for which y₀ and y'₀ are pure capacitive susceptances. The capacitive gyrator is used to produce the supercapacitor, or f.d.n.r., and the results of measurements on a number of f.d.n.r. networks are given.     

Eliciting information about organizational culture via laddering

Abstract. Eliciting information about organizational culture is an important part of system analysis and design. However, eliciting knowledge of this sort is difficult. Laddering is an established technique that is particularly suitable for eliciting information about goals and for eliciting explanations, which are important issues when investigating organizational culture. This paper describes the method, its strengths and limitations, its use in several case studies and its relation to other elicitation techniques. Recommendations for further work are given.     

Hemispheric asymmetries for global and local visual perception: Effects of stimulus and task factors.

Although neurological and physiological studies indicate a right hemisphere superiority in global processing and a left hemisphere superiority in local processing of Navon-type hierarchical letters (D. Navon, 1977), most investigations of lateralized perception in healthy participants report neither asymmetry. In 6 experiments the authors examined the influence of attentional demands, stimulus properties, and mode of response on perceptual asymmetries for global and local perception. Consistent with their theoretical predictions, asymmetries were more robust on divided- than focused-attention tasks and in response to stimuli in which local and global levels were equally salient compared with those with greater global than local saliency. Contrary to their prediction, perceptual asymmetries were not influenced by the complexity of the motor response. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

How initial conditions affect generalization performance in large networks.

Generalization is one of the most important problems in neural-network research. It is influenced by several factors in the network design, such as network size, weight decay factor, and others. We show here that the initial weight distribution (for gradient decent training algorithms) is one other factor that influences generalization. The initial conditions guide the training algorithm to search particular places of the weight space. For instance small initial weights tend to result in low complexity networks, and therefore can effectively act as a regularization factor. We propose a novel network complexity measure, which is helpful in shedding insight into the phenomenon, as well as in studying other aspects of generalization.     

Simpler Reasoning About System Properties: a Proof-by-Refinement Technique

Proofs about system specifications are difficult to conduct, particularly for large specifications. Using abstraction and refinement, we propose a proof technique that simplifies these proofs. We apply the technique to Circus (a combination of Z and CSP) specifications of different complexities. Interestingly, all the proofs are conducted in Z, even those concerning reactive behaviour.