Online State–Space Modeling Using Recurrent Multilayer Perceptrons with Unscented Kalman Filter

A nonlinear black-box modeling approach using a state–space recurrent multilayer perceptron (RMLP) is considered in this paper. The unscented Kalman filter (UKF), which was proposed recently and is appropriate for state–space representation, is employed to train the RMLP. The UKF offers a derivative-free computation and an easy implementation, compared to the extended Kalman filter (EKF) widely used for training neural networks. In addition, the UKF has a fast convergence rate and an excellent capability of parameter estimation which are appropriate for online learning. Through modeling experiments of nonlinear systems, the effectiveness of the RMLP trained with the UKF is demonstrated.     

Automatic recognition of vigilance state by using a wavelet-based artificial neural network

In this study, 5-s long sequences of full-spectrum electroencephalogram (EEG) recordings were used for classifying alert versus drowsy states in an arbitrary subject. EEG signals were obtained from 30 healthy subjects and the results were classified using a wavelet-based neural network. The wavelet-based neural network model, employing the multilayer perceptron (MLP), was used for the classification of EEG signals. A multilayer perceptron neural network (MLPNN) trained with the Levenberg–Marquardt algorithm was used to discriminate the alertness level of the subject. In order to determine the MLPNN inputs, spectral analysis of EEG signals was performed using the discrete wavelet transform (DWT) technique. The MLPNN was trained, cross-validated, and tested with training, cross-validation, and testing sets, respectively. The correct classification rate was 93.3% alert, 96.6% drowsy, and 90% sleep. The classification results showed that the MLPNN trained with the Levenberg–Marquardt algorithm was effective for discriminating the vigilance state of the subject.     

Study of symmetric microstructures for CMOS multilayer residual stress

This study presents a fabrication-based approach to improve the curl-up effect in complementary metal oxide semiconductor (CMOS) multilayer large-area planar structures. Control of the residual stress of CMOS multilayer microstructures is necessary for development of microelectromechanical systems (MEMS) sensors such as accelerometers and micromirrors. In this work, 3D symmetric geometry can be used to overcome effectively the residual stresses in CMOS multilayer microstructures. To demonstrate this concept, a symmetric multilayer flat-plane is fabricated and release-etched using an isotropic plasma etching process. The isotropic etch characteristics and lateral undercut can be controlled using a chamber pressure of 0.47 ± 0.2 Torr. A flat-plane structure with an area of 500 μm × 500 μm is fabricated using multilayer materials, including four metal and three silicon dioxide layers. Based on this approach, the measured results show the residual stress effect can be minimized in CMOS multilayer microstructures, and furthermore the curl-up effect of flat-plane is less than 2 μm across the 500 μm × 500 μm area.     

Effects of internal electrode composition on the reliability of low-firing PMN–PZT multilayer ceramic actuators

We investigated the effects of internal electrode composition on the reliability of low-firing multilayer ceramic actuators using Ag internal electrodes. Ag–ceramic composite pastes were prepared by adding Pb(Mg_1/3Nb_2/3)O₃–Pb(Zr_0.475,Ti_0.525)O₃ (PMNZT) ceramic powders to a commercial Ag paste at concentrations in the range of 0–60 vol%. PMNZT multilayered laminates were fabricated using tape casting, and then cofired at 925 °C for 10 h. The fatigue behaviors of multilayer actuators with Ag internal electrodes having different PMNZT concentrations were compared by applying a 2 kV/mm ac electric field at 50 °C under a relative humidity of 30%. The failure data were analyzed using Weibull statistics. The addition of PMNZT ceramics enhanced the mean time to failure by reducing the densification mismatch between the piezoelectric ceramic and internal electrode layers during the cofiring process.     

Numerical simulation of stochastic replicator models in catalyzed RNA-like polymers

A stochastic model for replicators in catalyzed RNA-like polymers is presented and numerically solved. The model consists of a system of reaction–diffusion equations describing the evolution of a population formed by RNA-like molecules with catalytic capabilities in a prebiotic process. The diffusion effects and the catalytic reactions are deterministic. A stochastic excitation with additive noise is introduced as a force term. To numerically solve the governing equations we apply the stochastic method of lines. A finite-difference reaction–diffusion system is constructed by discretizing the space and the associated stochastic differential system is numerically solved using a class of stochastic Runge–Kutta methods. Numerical experiments are carried out on a prototype of four catalyzed selfreplicator species along with an activated and an inactivated residues. Results are given in two space dimensions.     

Reinforcement Learning Using the Stochastic Fuzzy Min–Max Neural Network

The fuzzy min–max neural network constitutes a neural architecture that is based on hyperbox fuzzy sets and can be incrementally trained by appropriately adjusting the number of hyperboxes and their corresponding volumes. An extension to this network has been proposed recently, that is based on the notion of random hyperboxes and is suitable for reinforcement learning problems with discrete action space. In this work, we elaborate further on the random hyperbox idea and propose the stochastic fuzzy min–max neural network, where each hyperbox is associated with a stochastic learning automaton. Experimental results using the pole balancing problem indicate that the employment of this model as an action selection network in reinforcement learning schemes leads to superior learning performance compared with the traditional approach where the multilayer perceptron is employed.     

Combined projection and kernel basis functions for classification in evolutionary neural networks

This paper proposes a hybrid neural network model using a possible combination of different transfer projection functions (sigmoidal unit, SU, product unit, PU) and kernel functions (radial basis function, RBF) in the hidden layer of a feed-forward neural network. An evolutionary algorithm is adapted to this model and applied for learning the architecture, weights and node typology. Three different combined basis function models are proposed with all the different pairs that can be obtained with SU, PU and RBF nodes: product–sigmoidal unit (PSU) neural networks, product–radial basis function (PRBF) neural networks, and sigmoidal–radial basis function (SRBF) neural networks; and these are compared to the corresponding pure models: product unit neural network (PUNN), multilayer perceptron (MLP) and the RBF neural network. The proposals are tested using ten benchmark classification problems from well known machine learning problems. Combined functions using projection and kernel functions are found to be better than pure basis functions for the task of classification in several datasets.     

A symbolic approach to automatic multiword term structuring

This paper presents a three-level structuring of multiword terms basing on lexical inclusion, WordNet similarity and a clustering approach. Term clustering by automatic data analysis methods offers an interesting way of organizing a domain’s knowledge structure, useful for several information-oriented tasks like science and technology watch, textmining, computer-assisted ontology population, Question Answering (Q–A). This paper explores how this three-level term structuring brings to light the knowledge structures from a corpus of genomics and compares the mapping of the domain topics against a hand-built ontology (the GENIA ontology). Ways of integrating the results into a Q–A system are discussed.     

An implementation of the Expectation-Maximisation (EM) algorithm for population pharmacokinetic–pharmacodynamic modelling in ACSLXTREME

An implementation of the Expectation-Maximisation (EM) algorithm in ACSLXTREME (AEGIS Technologies) for the analyses of population pharmacokinetic–pharmacodynamic (PKPD) data is demonstrated. The parameter estimation results are compared with those from NONMEM (Globomax) using the first order conditional estimate method. The estimates are comparable and it is concluded that the EM algorithm is a useful technique in population pharmacokinetic–pharmacodynamic modelling. The implementation also demonstrates the ease with which parameter estimation algorithms for population data can be implemented in simulation software packages.     

Microfabrication of an electromagnetic power relay using SU-8 based UV-LIGA technology

Most of the MEMS relays reported in the field now are based on silicon fabrication and cannot be used for power applications. In this paper, we report a research effort to microfabricate an electromagnetic relay for power applications using a multilayer UV-LIGA process. A mechanically wrapped coil was used and very simple design for the magnetic circuit was adopted to increase the design flexibility and performances. The broad material selection and the capability of making high aspect ratio microstructures of the UV-LIGA make the technology best suited for fabricating microelectromechanical relays for power applications. The prototype relay has a truly three-dimensional structure and very suitable for large power capacity applications.The research work reported in this paper was made possible by support from National Science Foundation under grant ECS-#0104327; Louisiana Board of Regents, National Aeronautics and Space Administration, and the Louisiana Space Consortium under agreement NASA/LEQSF (2001–2005)-LaSPACE, and NASA/LaSPACE under grant NTG5–40115. The authors would also like to thank the Center for Advanced Microstructures and Devices of LSU for the use of the cleanroom facility.     

Dispersion of single-walled carbon nanotubes in poly(diallyldimethylammonium chloride) for preparation of a glucose biosensor

Poly(diallyldimethylammonium chloride) (PDDA) was chosen to disperse single-walled carbon nanotubes (SWCNTs). The optimal conditions to prepare stable PDDA–SWCNTs aqueous dispersions were presented. Then, the positively charged PDDA–SWCNTs composite and negatively charged glucose oxidase (GOD) were employed to fabricate multilayer films on platinum (Pt) electrodes by layer-by-layer self-assembly technique. The consecutive growth of the multilayer films was confirmed by quartz crystal microbalance. Electrochemical measurements were used to study the properties of the proposed biosensor. Results demonstrated that SWCNTs were evenly dispersed within the PDDA films and efficiently improved the conductivity of the resulting films. Among the biosensors, the one based on seven layers of multilayer films got the best performance. It showed wide linear range of 0.05–12 mM, high sensitivity of 63.84 μA/(mM cm²), low detection limit of about 4 μM and small value of the apparent Michaelis–Menten constant, 8.46 mM. In addition, the biosensor also exhibited good suppression of interference and long-term operational stability. This protocol could be used to immobilize other enzymes to construct a range of biosensors.     

Efficient Minimisation of the KL Distance for the Approximation of Posterior Conditional Probabilities

The minimisation of a least mean squares cost function produces poor results in the ranges of the input variable where the quantity to be approximated takes on relatively low values. This can be a problem if an accurate approximation is required in a wide dynamic range. The present paper approaches this problem in the case of multilayer perceptrons trained to approximate the posterior conditional probabilities in a multicategory classification problem. The use of a cost function derived from the Kullback–Leibler information distance measure is proposed and a computationally light algorithm is derived for its minimisation. The effectiveness of the procedure is experimentally verified.     

A mathematical model of evolution of biological populations

We determine the laws of evolution of closed populations in a self-regulating biogeocenosis. We show that the evolutionary moving force creating new genotypes is the variability of the biotypic effects, whereas the evolutionary factor that regulates (stabilizes) the size of the population is the variability of the biocenotic effects.Translated from Kibernetika, No. 1, pp. 107–111, January–February, 1990.     

Multilayer Potts Perceptrons With Levenberg–Marquardt Learning

This paper presents learning multilayer Potts perceptrons (MLPotts) for data driven function approximation. A Potts perceptron is composed of a receptive field and a $K$ -state transfer function that is generalized from sigmoid-like transfer functions of traditional perceptrons. An MLPotts network is organized to perform translation from a high-dimensional input to the sum of multiple postnonlinear projections, each with its own postnonlinearity realized by a weighted $K$-state transfer function. MLPotts networks span a function space that theoretically covers network functions of multilayer perceptrons. Compared with traditional perceptrons, weighted Potts perceptrons realize more flexible postnonlinear functions for nonlinear mappings. Numerical simulations show MLPotts learning by the Levenberg–Marquardt (LM) method significantly improves traditional supervised learning of multilayer perceptrons for data driven function approximation.      

Ab initio calculation of the BCC Fe–Al–Mo (Iron–Aluminum–Molybdenum) phase diagram: Implications for the nature of the phase

The metastable phase diagram of the BCC-based ordering equilibria in the Fe–Al–Mo system has been calculated via a truncated cluster expansion, through the combination of Full-Potential-Linear augmented Plane Wave (FP-LAPW) electronic structure calculations and of Cluster Variation Method (CVM) thermodynamic calculations in the irregular tetrahedron approximation. Four isothermal sections at 1750 K, 2000 K, 2250 K and 2500 K are calculated and correlated with recently published experimental data on the system. The results confirm that the critical temperature for the order–disorder equilibrium between Fe₃Al–D0₃ and FeAl–B2 is increased by Mo additions, while the critical temperature for the FeAl–B2/A2 equilibrium is kept approximately invariant with increasing Mo contents. The stabilization of the Al-rich A2 phase in equilibrium with overstoichiometric B2–(Fe,Mo)Al is also consistent with the attribution of the A2 structure to the τ₂ phase, stable at high temperatures in overstoichiometric B2–FeAl.     

Control of particle size distribution described by a population balance model of semibatch emulsion polymerization

This work addresses the computation of surfactant feed profiles for shaping a particle size distribution to a bimodal target distribution. A fundamental population balance model of styrene semibatch emulsion polymerization is used in this control study. Both surfactant feed rate and free surfactant concentration are considered as alternative control variables and a comparison is made of the two approaches. A comparison is also made of several different objective function norms in the optimization. Results suggest that a min–max norm, tied to the distribution modes, is the most appropriate metric.     

A Genetic Algorithm Simulation of a Transition Economy: An Application to Insider-Privatization in Croatia

A genetic algorithm simulation is applied to a model of privatization in a transition economy. Bounded-rational agents, learning by doing in a changing economic environment, are presented as a population of artificial adaptive agents. The paper examines the comparative performances of three alternative forms of genetic algorithm – the simple GA, PGA with election and EGA with elite selection. The latter version proved to be more robust than the alternatives.     

Size optimization of space trusses using Big Bang–Big Crunch algorithm

A Hybrid Big Bang–Big Crunch (HBB–BC) optimization algorithm is employed for optimal design of truss structures. HBB–BC is compared to Big Bang–Big Crunch (BB–BC) method and other optimization methods including Genetic Algorithm, Ant Colony Optimization, Particle Swarm Optimization and Harmony Search. Numerical results demonstrate the efficiency and robustness of the HBB–BC method compared to other heuristic algorithms.     

ISE-optimal nonminimum-phase compensation for nonlinear processes

This work concerns the optimal regulation of single-input–single-output nonminimum-phase nonlinear processes. The problem of calculation of an ISE-optimal, statically equivalent, minimum-phase output for nonminimum-phase compensation is formulated using Hamilton–Jacobi theory and the normal form representation of the nonlinear system. A Newton–Kantorovich iteration is developed for the solution of the pertinent Hamilton–Jacobi equations, which involves solving a Zubov equation at each step of the iteration. The method is applied to the problem of controlling a nonisothermal CSTR with Van de Vusse kinetics, which exhibits nonminimum-phase behaviour.     

Construction of modified embedded atom method potentials for the study of the bulk phase behaviour in binary Pt–Rh, Pt–Pd, Pd–Rh and ternary Pt–Pd–Rh alloys

A first attempt is made to simulate the solid part of the phase diagram of the ternary Pt–Pd–Rh system. To this end, Monte Carlo (MC) simulations are combined with the Modified Embedded Atom Method (MEAM) and optimised parameters entirely based on Density Functional Theory (DFT) data. This MEAM potential is first validated by calculating the heat of mixing or the demixing phase boundary for the binary subsystems Pt–Rh, Pt–Pd and Pd–Rh. For the disordered alloy systems Pt–Rh and Pt–Pd, the MC/MEAM simulation results show a slightly exothermic heat of mixing, thereby contradicting any demixing behaviour, in agreement with other theoretical results. For the Pd–Rh system the experimentally observed demixing region is very well reproduced by the MC/MEAM simulations. The extrapolation of the MEAM potentials to ternary systems is next validated by comparing DFT calculations for the energy of formation of ordered Pt–Pd–Rh compounds with the corresponding MEAM energies. Finally, the validated potential is used for the calculation of the ternary phase diagram at 600 K.