Evolving Stochastic Learning Algorithm based on Tsallis entropic index

In this paper, inspired from our previous algorithm, which was based on the theory of Tsallis statistical mechanics, we develop a new evolving stochastic learning algorithm for neural networks. The new algorithm combines deterministic and stochastic search steps by employing a different adaptive stepsize for each network weight, and applies a form of noise that is characterized by the nonextensive entropic index q, regulated by a weight decay term. The behavior of the learning algorithm can be made more stochastic or deterministic depending on the trade off between the temperature T and the q values. This is achieved by introducing a formula that defines a time-dependent relationship between these two important learning parameters. Our experimental study verifies that there are indeed improvements in the convergence speed of this new evolving stochastic learning algorithm, which makes learning faster than using the original Hybrid Learning Scheme (HLS). In addition, experiments are conducted to explore the influence of the entropic index q and temperature T on the convergence speed and stability of the proposed method.     

Vapor liquid equilibrium modeling of alkane systems with Equations of State: “Simplicity versus complexity”

Two types of Equations of State (EoS), which are characterized here as “simple” and “complex” EoS, are evaluated in this study. The “simple” type involves two versions of the Peng–Robinson (PR) EoS: the traditional one that utilizes the experimental critical properties and the acentric factor and the other, referred to as PR-fitted (PR-f), where these parameters are determined by fitting pure compound vapor pressure and saturated liquid volume data. As “complex” EoS in this study are characterized the EoS derived from statistical mechanics considerations and involve the Sanchez–Lacombe (SL) EoS and two versions of the Statistical Associating Fluid Theory (SAFT) EoS, the original and the Perturbed-Chain SAFT (PC-SAFT).

The evaluation of these two types of EoS is carried out with respect to their performance in the prediction and correlation of vapor liquid equilibria in binary and multicomponent mixtures of methane or ethane with alkanes of various degree of asymmetry. It is concluded that for this kind of systems complexity offers no significant advantages over simplicity. Furthermore, the results obtained with the PR-f EoS, especially those for multicomponent systems that are encountered in practice, even with the use of zero binary interaction parameters, indicate that this EoS may become a powerful tool for reservoir fluid phase equilibria modeling.     

Structural parameters of the ground states of the quasi‐stable diatomic anions CO−, BF−, and BCl− as obtained by conventional Ab Initio methods

The experimental electron affinity (EA) of CO(X¹Σ⁺) is ?1.5 eV, signifying the metastability of the CO^?(X²Π) anion. The electronic structure and bonding of CO^?, BF^?, and BCl^? vis‐à‐vis their neutral counterparts have been studied by conventional coupled‐cluster (CCSD(T)) and multireference (MRCI) methods. Our results are in agreement with experiment for the CO/CO^? system, indicating as well the metastable nature of the BF^?(X²Π) and BCl^?(X²Π) anions, their MRCI EAs being ?0.8 ± 0.1 and ?0.3 ± 0.1 eV, respectively. Our work clearly shows the usefulness of stationary state ab initio methods to the elucidation of metastable species. © 2015 Wiley Periodicals, Inc.     

Nonmonotone BFGS-trained recurrent neural networks for temporal sequence processing

In this paper we propose a nonmonotone approach to recurrent neural networks training for temporal sequence processing applications. This approach allows learning performance to deteriorate in some iterations, nevertheless the network’s performance is improved over time. A self-scaling BFGS is equipped with an adaptive nonmonotone technique that employs approximations of the Lipschitz constant and is tested on a set of sequence processing problems. Simulation results show that the proposed algorithm outperforms the BFGS as well as other methods previously applied to these sequences, providing an effective modification that is capable of training recurrent networks of various architectures.     

Layer potential techniques for the narrow escape problem

The narrow escape problem consists in deriving the asymptotic expansion of the solution of a drift-diffusion equation with the Dirichlet boundary condition on a small absorbing part of the boundary and the Neumann boundary condition on the remaining reflecting boundaries. Using layer potential techniques, we rigorously find high-order asymptotic expansions of such solutions. The asymptotic formula explicitly exhibits the nonlinear interaction of many small absorbing targets. Based on the asymptotic theory for eigenvalue problems developed in Ammari et al. (2009) [3], we also construct high-order asymptotic formulas for the perturbation of eigenvalues of the Laplace and the drifted Laplace operators for mixed boundary conditions on large and small pieces of the boundary.     

Prediction Helps Analytical Experimental Work for Environmental Purposes

 Analytical work, especially for environmental purposes, involves in the typical case determination of organic compounds in compartments, where their presence is very small. Identification of the appropriate experimental technique, whose range of applicability covers the concentration value involved, requires an estimate of this value. To this purpose, we present in this study a methodology for the prediction of the concentration of organic pollutants in the various environmental compartments (aquatic biota, air, sediment and soil) from the knowledge of the concentration in one of them. In case where the pollutant’s concentration is not available for any of the compartments, the proposed methodology can be used to estimate the maximum expected concentration in each one of them assuming that the water phase is saturated with the pollutant. The latter value can be obtained from a simple correlation presented here. The methodology is based on the correlation of the partition coefficients of pollutants among the various environmental compartments with two important thermodynamic quantities of pollutants: the octanol/water partition coefficient and the Henry’s law coefficient, which can be easily predicted by simple models presented here. Application of the methodology to field experimental data gives very satisfactory results at least for identifying the appropriate experimental technique.     

Design,Synthesis and Antiparasitic Evaluation of Click Phospholipids

A library of seventeen novel ether phospholipid analogues, containing 5-membered heterocyclic rings (1,2,3-triazolyl, isoxazolyl, 1,3,4-oxadiazolyl and 1,2,4-oxadiazolyl) in the lipid portion were designed and synthesized aiming to identify optimised miltefosine analogues. The compounds were evaluated for their in vitro antiparasitic activity against Leishmania infantum and Leishmania donovani intracellular amastigotes, against Trypanosoma brucei brucei and against different developmental stages of Trypanosoma cruzi. The nature of the substituents of the heterocyclic ring (tail) and the oligomethylene spacer between the head group and the heterocyclic ring was found to affect the activity and toxicity of these compounds leading to a significantly improved understanding of their structure–activity relationships. The early ADMET profile of the new derivatives did not reveal major liabilities for the potent compounds. The 1,2,3-triazole derivative 27 substituted by a decyl tail, an undecyl spacer and a choline head group exhibited broad spectrum antiparasitic activity. It possessed low micromolar activity against the intracellular amastigotes of two L. infantum strains and T. cruzi Y strain epimastigotes, intracellular amastigotes and trypomastigotes, while its cytotoxicity concentration (CC₅₀) against THP-1 macrophages ranged between 50 and 100 μM. Altogether, our work paves the way for the development of improved ether phospholipid derivatives to control neglected tropical diseases.     

Microfluidic device to investigate factors affecting performance in biosensors designed for transdermal applications

In this work we demonstrate a novel microfluidic based platform to investigate the performance of 3D out-of-plane microspike array based glucose and lactate biosensors. The microspike array was bonded with a glass slide and modified with glucose oxidase or lactate oxidase using covalent coupling chemistry. An epoxy-polyurethane based membrane was used to extend the linear working range (from 0 to 25 mM of substrate) of these biosensors. Both lactate and glucose sensors performed well in the clinically relevant substrate concentration range. Glucose microspikes were further investigated with respect to the effects of substrate transfer by incorporation into a microfluidic system. Data from the microfluidic system revealed that the sensor response is mainly dependent on enzyme kinetics rather than membrane permeability to glucose. The robustness of the sensors was demonstrated by its consistency in performance extending over 48 h.     

Omnidirectional antireflective properties of porous tungsten oxide films with in-depth variation of void fraction and stoichiometry

We report on the fabrication of porous hot-wire deposited WO_x (hwWO_x) films with omnidirectional antireflective properties coming from in-depth variation of both (i) void fraction from 0% at the Si substrate/hwWO_x interface to 30% within less than 7 nm and to higher than 50% at the hwWO_x/air interface, and (ii) x, namely hwWO_x stoichiometry, from 2.5 at the Si/hwWO_x to 3 within less than 7 nm. hwWO_x films were deposited by means of hw deposition at rough vacuum and controlled chamber environment. The films were analyzed by Spectroscopic Ellipsometry to extract the graded refractive index profile, which was then used in a rigorous coupled wave analysis (RCWA) model to simulate the antireflective properties. RCWA followed reasonably the experimental reflection measurements. Void fraction and x in-depth variation, controlled by the hw process, greatly affect the antireflective properties, and improve the omnidirectional and broadband characteristics. The reflection suppression below 10% within the range of 500-1000 nm for angles of incidence up to more than 60° is demonstrated.