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.     

Influence of two-step process and of different stoichiometric ratios on morphologies generated in castor oil epoxy resins

Phase separation during polymerization was studied in a model system consisting of a diepoxide based on diglycidyl ether of bisphenol A (DGEBA), variable amounts of ethylenediamine (EDA) and the mass of castor oil (CO) necessary to obtain a mass fraction equal to 0-15 in a final system where the stoichiometric ratio of amine to epoxy equivalents, r, was equal to 1. A two-step polymerization process was performed by curing first a system with r = 0-5, during variable times before phase separation, and then carrying the system to r = 1. Thermodynamic analysis of samples with different r values led to a linear relationship between the Flory-Huggins interaction parameter and r. The concentration (P) and average size (D?) of dispersed-phase particles followed opposite trends, i.e. P increased while D? decreased, when either r was increased or the time of curing in the first step of a two-step process was decreased. This was explained by assuming that the competition between nucleation and growth was determined by the viscosity at the cloud point, η_cp. Low values of η_cp favoured growth over nucleation and led to fewer but larger particles.     

Crosslinking of tissue-derived biomaterials in 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC)

In contrast to bifunctional reagents such as glutaraldehyde or polyfunctional reagents such as polyepoxides, carbodiimides belong to the class of zero-length crosslinkers which modify amino acid side-groups to permit crosslink formation, but do not remain as part of that linkage. The authors have compared the effects of 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC) and glutaraldehyde (the de facto industrial standard crosslinker) on the hydrothermal, biochemical, and uniaxial mechanical properties of bovine pericardium. EDC crosslinking was optimized for maximum increase in collagen denaturation temperature using variables of pH, concentration, and ratio of EDC to N-hydroxysuccinimide (NHS): a reagent for formation of activated esters. EDC and glutaraldehyde crosslinked materials were subjected to hydrothermal denaturation tests, biochemical degradation by enzymes (collagenase, trypsin) and CNBr, amino acid analysis for unreacted lysine, and to high strain rate mechanical tests including: large deformation stress-strain studies (0.1 to 10 Hz), stress relaxation experiments (loading time 0.1 s) and small deformation forced vibration (1 and 10 Hz). A protocol for EDC crosslinking was developed which used 1.15% EDC (2:1 EDC:NHS) at pH 5.5 for 24 h. The increase in denaturation temperature for EDC (from 69.7±1.2°C to 86.0±0.3°C) was equivalent to that produced by glutaraldehyde (85.3±0.4°C). Both treatments equivalently increased resistance to collagenase and CNBr degradation; however, after denaturation, the EDC-treated tissue was slightly more resistant to collagenase, and markedly more resistant to trypsin. EDC-treated materials were more extensible and more elastic than glutaraldehyde-treated materials. Despite the differences in crosslinking mechanism, EDC and glutaraldehyde-treated materials are very similar. Subtle but intriguing differences in biochemical structure remain to be investigated.     

Global buckling load of a nonlinearly elastic bar

Summary The nonlinear buckling of an elastic bar with softening characteristic is studied. It is found that, under large disturbances, the bar may buckle under loads well below the linear buckling load.     

Specific heat of a marginal Fermi liquid

The temperature dependence of a marginal Fermi liquid has been calculated. We showed that the expectedT lnT correction is characteristic for the low-temperature domain. The high-temperature domain has a supplementary correction. The results are in agreement with the non-Fermi behavior of some metallic systems the in low-temperature domain.     

Beam test of the CsI(Tl) calorimeter for the BELLE detector at the KEK-B factory

We studied the performance of a prototype electromagnetic calorimeter for the BELLE detector at the KEK proton synchrotron for an energy range of 0.25–3.5 GeV. The prototype consisted of an array of 6 × 5 CsI(Tl) crystals with 30 cm length (16.2 radiation lengths) and about 6 cm × 6 cm cross section. The scintillation light of each CsI(Tl) crystal was read out by two large-area PIN photodiodes and charge-sensitive preamplifiers attached at the rear face of the crystal. We measured the energy and position resolution for electrons and the e/π separation for two sets of matrix configurations: one corresponded to the center and the other to the edge of the barrel calorimeter. The overall performance measured by the test proves that the prototype calorimeter is satisfactory for the use in the BELLE detector.     

High performance CMOS current comparator

A new high-speed CMOS current comparator is described. By changing the buffer of an existing comparator from class B to class AB operation, voltage swings are reduced, resulting in greater speed for small input currents. Simulation results employing a 1.6μm CMOS technology show the response time to a 0.1μA input current to be 11ns and the power dissipation to be 1.4mW, resulting in a five times improvement in speed/power ratio over existing high-speed current comparators