A global optimum approach for one-layer neural networks

The article presents a method for learning the weights in one-layer feedforward neural networks minimizing either the sum of squared errors or the maximum absolute error, measured in the input scale. This leads to the existence of a global optimum that can be easily obtained solving linear systems of equations or linear programming problems, using much less computational power than the one associated with the standard methods. Another version of the method allows computing a large set of estimates for the weights, providing robust, mean or median, estimates for them, and the associated standard errors, which give a good measure for the quality of the fit. Later, the standard one-layer neural network algorithms are improved by learning the neural functions instead of assuming them known. A set of examples of applications is used to illustrate the methods. Finally, a comparison with other high-performance learning algorithms shows that the proposed methods are at least 10 times faster than the fastest standard algorithm used in the comparison.     

Modeling of a coil system considering frequency-dependent inductances and losses I. Analysis of the impedance matrix characteristics

 A general equivalent circuit for the modeling of a system of magnetically coupled coils including the losses is presented. This equivalent circuit includes two parts: a subsystem of primary branches whose terminals correspond with those of the real coils, and a secondary network which allows the representation of frequency-dependent inductances and resistances. An analysis of the equations of the proposed circuit is carried out and an expression of the frequency-dependent impedance matrix as seen by the primary coils is determined. The properties of the mathematical expression are analyzed. The proposed equivalent circuit allows representing the dependence of inductances and resistances on the frequency. The expression of the impedance is the starting point for the synthesis of an equivalent circuit, which is carried out in a separate article. Received: 25 May 2001/Accepted: 25 June 2001     

The Effect of ceramic reinforcements during spray atomization and codeposition of metal matrix composites: Part II. Solid-state cooling effects

In the present study, the mechanisms governing the evolution of microstructure during spray atomization and codeposition of metal matrix composites (MMCs) were investigated, with particular emphasis on the effects of the ceramic phase on the resulting microstructure during solidstate cooling. The grain size refinement that is commonly observed when a distribution of ceramic particulates is coinjected into a metallic spray during spray atomization and deposition processing was rationalized in terms of three distinct effects: (a) solidification effects, (b) heattransfer effects, and (c) solid-state cooling effects. The solidification and heat-transfer effects were discussed in Part I. [1] Regarding solid-state cooling effects, the present results show that the presence of a dispersion of ceramic particulates in the aluminum matrix effectively reduces the rate of grain growth during solid-state cooling. Experimental support to this suggestion was provided by the results of an investigation on the changes in grain size following isochronal thermal anneals.     

Interaction mechanisms between ceramic particles and atomized metallic droplets

The present study was undertaken to provide insight into the dynamic interactions that occur when ceramic particles are placed in intimate contact with a metallic matrix undergoing a phase change. To that effect, Al-4 wt pct Si/SiC_p composite droplets were synthesized using a spray atomization and coinjection approach, and their solidification microstructures were studied both qualitatively and quantitatively. The present results show that SiC particles (SiC_p) were incor- porated into the matrix and that the extent of incorporation depends on the solidification con- dition of the droplets at the moment of SiC particle injection. Two factors were found to affect the distribution and volume fraction of SiC particles in droplets: the penetration of particles into droplets and the entrapment and/or rejection of particles by the solidification front. First, during coinjection, particles collide with the atomized droplets with three possible results: they may penetrate the droplets, adhere to the droplet surface, or bounce back after impact. The extent of penetration of SiC particles into droplets was noted to depend on the kinetic energy of the particles and the magnitude of the surface energy change in the droplets that occurs upon impact. In liquid droplets, the extent of penetration of SiC particles was shown to depend on the changes in surface energy, ΔE_s, experienced by the droplets. Accordingly, large SiC particles encoun- tered more resistance to penetration relative to small ones. In solid droplets, the penetration of SiC particles was correlated with the dynamic pressure exerted by the SiC particles on the droplets during impact and the depth of the ensuing crater. The results showed that no pene- tration was possible in such droplets. Second, once SiC particles have penetrated droplets, their final location in the microstructure is governed by their interactions with the solidification front. As a result of these interactions, both entrapment and rejection of SiC particles occurred during droplet solidification. A comparison of the present results to those anticipated from well-established kinetic and thermodynamic models led to some interesting findings. First, the models proposed by Boiling and Cisse^[24] and Chernovet al.^[58] predict relative low critical interface velocities necessary for entrapment, inconsistent with the present experimental findings. Second, although the observed correlation between the critical front velocity and droplet diameter was generally consistent with that predicted by Stefanescuet a/.’s model,^[27] the dependence on the size of SiC particles was not. In view of this discrepancy, three possible mechanisms were proposed to account for the experimental findings: nucleation of α-Al on SiC particles, entrapment of SiC particles between primary dendrite arms, and entrapment of SiC particles between secondary dendrite arms.     

Supply Estimation Using Coevolutionary Genetic Algorithms in the Spanish Electrical Market

The price of electrical energy in Spain has not been regulated by the government since 1998, but determined by the supply from the generators in a competitive market, the so-called electrical pool. A genetic method for analyzing data from this new market is presented in this paper. The eventual objective is to determine the individual supply curves of the competitive agents. Adopting the point of view of the game theory, different genetic algorithm configurations using coevolutionary and non-coevolutionary strategies combined with scalar and multi-objective fitness are compared. The results obtained are the first step toward solving the induction of the optimal individual strategies into the Spanish electrical market from data in terms of perfect oligopolistic behavior.     

Temporal transformation of periodic incoherent ultrashort light pulses by chirped fiber gratings

The analogy between free-space propagation of optical beams and light-pulse reflection from linearly chirped fiber gratings is used to analyze the Lau effect in the temporal domain. The coherence conditions that are satisfied in the spatial domain for obtaining, at certain fixed locations, periodic fringes patterns are reformulated for guided light propagation. In this analogy, spatial periodic irradiance distributions are transformed in periodic sequences of light pulses. An optical setup is proposed to produce sharp pulse trains, with minimal distortion effects, that have repetition frequencies that are different from those associated with the input periodic optical signal. Some numerical results are given to illustrate this approach.     

New learning rules for the ASSOM network

The ASSOM is a self-organising neural network with the capability of adapting to linear subspaces. Here we propose two new methods to train the ASSOM network. A nonlinear system of equations is derived for network training. This system can be solved by a gradient-based approach or by the Levenberg–Marquardt method. Each of these two approaches gives a different learning rule. A comparison is carried out among the original Kohonens method and the proposed learning rules. Experimental results are reported, including a convergence speed experiment and a speech processing application, which show that the new learning rules have better performance than the original one.     

AC-coupled front-end for biopotential measurements

AC coupling is essential in biopotential measurements. Electrode offset potentials can be several orders of magnitude larger than the amplitudes of the biological signals of interest, thus limiting the admissible gain of a dc-coupled front end to prevent amplifier saturation. A high-gain input stage needs ac input coupling. This can be achieved by series capacitors, but in order to provide a bias path, grounded resistors are usually included, which degrade the common mode rejection ratio (CMRR). This paper proposes a novel balanced input ac-coupling network that provides a bias path without any connection to ground, thus resulting in a high CMRR. The circuit being passive, it does not limit the differential dc input voltage. Furthermore, differential signals are ac coupled, whereas common-mode voltages are dc coupled, thus allowing the closed-loop control of the dc common mode voltage by means of a driven-right-leg circuit. This makes the circuit compatible with common-mode dc shifting strategies intended for single-supply biopotential amplifiers. The proposed circuit allows the implementation of high-gain biopotential amplifiers with a reduced number of parts, thus resulting in low power consumption. An electrocardiogram amplifier built according to the proposed design achieves a CMRR of 123 dB at 50 Hz.     

Identification of protein ligands in complex biological samples using intensity-fading MALDI-TOF mass spectrometry

The easy detection of biomolecular interactions in complex mixtures using a minimum amount of material is of prime interest in molecular and cellular biology research. In this work, a mass spectrometry MALDI-TOF based approach, which we call intensity-fading (IF MALDI-TOFMS), and which was designed for just such a purpose, is reported. This methodology is based on the use of the MALDI ion intensities to detect quickly the formation of complexes between nonimmobilized biomolecules in which a protein is one of the partners (protein-protein, protein-peptide, protein-organic molecule, and protein-nucleic acid complexes). The complex is detected through the decrease (fading) of the molecular ion intensities of the partners as directly compared to the MALDI mass spectrum of the mixture (problem and control molecules) following the addition of the target molecule. The potential of the approach is examined in several examples of model interactions, mainly involving small nonprotein and protein inhibitors of proteases, at both the qualitative and semiquantitative levels. Using this method, different protein ligands of proteolytic enzymes in total extracts of invertebrate organisms have been identified in a simple way. The proposed procedure should be easily applied to the high-throughput screening of biomolecules, opening a new experimental strategy in functional proteomics.     

Structures and stabilities of small DNA dumbbells with Watson-Crick and Hoogsteen base pairs

The structures and stabilities of cyclic DNA octamers of different sequences have been studied by NMR and CD spectroscopy and by restrained molecular dynamics. At low oligonucleotide concentrations, some of these molecules form stable monomeric structures consisting of a short stem of two base pairs connected by two mini-loops of two residues. To our knowledge, these dumbbell-like structures are the smallest observed to date. The relative stabilities of these cyclic dumbbells have been established by studying their melting transitions. Dumbbells made up purely of GC stems are more stable than those consisting purely of AT base pairs. The order of the base pairs closing the loops also has an important effect on the stabilities of these structures. The NMR data indicate that there are significant differences between the solution structures of dumbbells with G-C base pairs in the stem compared to those with A-T base pairs. In the case of dumbbells with G-C base pairs, the residues in the stem form a short segment of a BDNA helix stabilized by two Watson-Crick base pairs. In contrast, in the case of d, the stem is formed by two A-T base pairs with the glycosidic angles of the adenine bases in a syn conformation, most probably forming Hoogsteen base pairs. Although the conformations of the loop residues are not very well defined, the thymine residues at the first position of the loop are observed to fold back into the minor groove of the stem.