A method for improving classification reliability of multilayerperceptrons

Criteria for evaluating the classification reliability of a neural classifier and for accordingly making a reject option are proposed. Such an option, implemented by means of two rules which can be applied independently of topology, size, and training algorithms of the neural classifier, allows one to improve the classification reliability. It is assumed that a performance function P is defined which, taking into account the requirements of the particular application, evaluates the quality of the classification in terms of recognition, misclassification, and reject rates. Under this assumption the optimal reject threshold value, determining the best trade-off between reject rate and misclassification rate, is the one for which the function P reaches its absolute maximum. No constraints are imposed on the form of P, but the ones necessary in order that P actually measures the quality of the classification process. The reject threshold is evaluated on the basis of some statistical distributions characterizing the behavior of the classifier when operating without reject option; these distributions are computed once the training phase of the net has been completed. The method has been tested with a neural classifier devised for handprinted and multifont printed characters, by using a database of about 300000 samples. Experimental results are discussed.     

The effect of support morphology on the reaction of oxidative dehydrogenation of ethane to ethylene at short contact times

The oxidative dehydrogenation of ethane is carried out in short contact time reactors over Pt and LaMnO₃ based catalysts supported on a large number of different ceramic substrates (45, 60 and 80 ppi foam monoliths and 200, 400, 600, 900 and 1200 cpsi honeycomb monoliths). Experimental results, obtained under the same conditions at varying the C₂H₆/O₂ ratio, showed that the highest performance in terms of ethylene selectivity and yield is always attained on LaMnO₃ catalysts. Furthermore, the results are significantly influenced by the morphology and cell density of the support, with 45 and 60 ppi foams and 400 and 600 cpsi honeycombs giving the best performance. The experimental results are explained by means of geometrical and fluid dynamic considerations on the support, and by means of a 2D mathematical model, which clearly indicates an optimal intermediate cell density for maximising ethylene selectivity and yield.     

Surface adsorption of comb polymers by Monte Carlo simulations

This paper reports off-lattice Monte Carlo simulations of highly-branched comb homopolymers weakly adsorbed on a flat, featureless surface showing only covolume and dispersion interactions with the adsorbate. A minimal coarse-grained model, described by hard spheres connected by harmonic springs, was employed. The interaction energy of the adsorbed combs and linear chains is first discussed as a function of the molecular mass and of the number of beads in contact with the surface. The molecular size is then investigated as a function of backbone length and branching density at a fixed arm size. The apparent swelling exponents of the adsorbed combs are larger than those of the corresponding linear chains, and much larger than that of the free molecules. This result indicates a surface-induced stiffening of the comb backbone, further studied through the persistence length l_pers. It is found that l_pers increases upon adsorption over the free-molecule value, more so the larger is the branching density. Finally, the thickness of the adsorbed layer, the surface-induced molecular anisotropy and the molecular aspect ratio are investigated as a function of branching density and molecular mass.     

Good teaching inputs do not correspond to desired responses in ecological neural networks

In this paper we analyze how supervised learning occurs in ecological neural networks, i.e. networks that interact with an autonomous external environment and, therefore, at least partially determine with their behavior their own input. Using an evolutionary method for selecting good teaching inputs we surprisingly find that to obtain a desired outputX it is better to use a teaching input different fromX. To explain this fact we claim that teaching inputs in ecological networks have two different effects: (a) to reduce the discrepancy between the actual output of the network and the teaching input, (b) to modify the network's behavior and, as a consequence, the network's learning experiences. Evolved teaching inputs appear to represent a compromise between these two needs. We finally show that evolved teaching inputs that are allowed to change during the learning process function differently at different stages of learning, first giving more weight to (b) and, later on, to (a).     

Quantification of secondary structure prediction improvement using multiple alignments

The use of multiple sequence alignments for secondary structurepredictions is analysed. Seven different protein families, containingonly sequences of known structure, were considered to providea range of alignment and prediction conditions. Using alignmentsobtained by spatial superposition of main chain atoms in knowntertiary protein structures allowed a mean of 8% in secondarystructure prediction accuracy, when compared to those obtainedfrom the individual sequences. Substitution of these alignmentsby those determined directly from an automated sequence alignmentalgorithm showed variations in the prediction accuracy whichcorrelated with the quality of the multiple alignments and distanceof the primary sequence. Secondary structure predictions canbe reliably improved using alignments from an automatic alignmentprocedure with a mean increase of 6.87percnt;, giving an overallprediction accuracy of 68.5%, if there is a minimum of 25% sequenceidentity between all sequences in a family.     

A voltammetric study concerning the structural stability of Li-overstoichiometric Mg-doped LiCoO2 powders

The effect of Mg-doping and Li overstoichiometry on the structural stability of LiCoO₂ powders has been investigated with emphasis to voltammetric properties. Microparticle cyclic voltammetry (CV) conducted in caustic NaOH to best simulate a non-aqueous electrolyte shows a marked improvement of the structure stability of doped LiCoO₂. In contrast to the unsubstituted LiCoO₂ sample which shows voltammetric peaks associated to the well-known two-phase domain and monoclinic distortion reactions, in Li_1.08Mg_0.06CoO₂, LiMg_0.06CoO₂ and Li_1.08CoO₂ samples these peaks are strongly suppressed providing direct evidence for the existence of a stable solid solution with negligible phase transitions in the reversible intercalation region (3.8-4.2 V vs. Li) as well as in the overcharged region. The effect is higher with Mg-doping, irrespective of the Li overstoichiometry. However, the concomitant presence of Mg and Li excess in the structure is important for obtaining small particle sizes. Since Mg-doping induces a quasi metallic behavior in the samples, whereas the Li excess may provide an higher initial capacity, it is suggested that the Li_1.08Mg_0.06CoO₂ composition may be of interest as positive cathode for advanced Li-ion batteries.     

Chemical and Physical Signals Mediating Conspecific and Heterospecific Aggregation Behavior of First Instar Stink Bugs

We investigated cues that mediate the aggregation behavior of immature pentatomid bugs by using nymphs of six different pentatomid bug species (Nezara viridula, Acrosternum hilare, Chlorochroa ligata, Chlorochroa sayi, Thyanta pallidovirens, and Euschistus conspersus). When first instars of any two species were put together in a Petri dish, they readily formed heterospecific aggregations similar to their natural conspecific aggregations. The chemical profiles of first and second instar nymphs of each species were determined by solvent extraction with pentane, followed by GC-MS analysis. Immature bugs of the different species had some compounds in common, and some that were more species specific. Within a species, there were distinct differences in the profiles of compounds extracted from first and second instars. Bugs did not aggregate around untreated polysulfone beads (1 mm diam) that were glued together in groups approximating bug egg masses, suggesting that tactile cues alone were insufficient to induce aggregation. Furthermore, when tested over a range of doses, groups of polysulfone beads treated with crude whole-body extracts of bugs did not induce or maintain aggregations. However, first instar N. viridula nymphs did respond to beads treated with two of the three major components of bug extracts. 4-Oxo-(E)-2-decenal induced significant aggregations at two doses, whereas tridecane, the major component in extracts from all six species, did not, and (E)-2-decenal was repellent. The repellence of (E)-2-decenal may explain why we and previous researchers were unable to induce aggregations of first instar N. viridula using whole-body extracts.     

Test Technology Newsletter

Journal of Electronic Testing -