Learning higher-level features with convolutional restricted Boltzmann machines for sentiment analysis

In recent years, learning word vector representations has attracted much interest in Natural Language Processing. Word representations or embeddings learned using unsupervised methods help addressing the problem of traditional bag-of-word approaches which fail to capture contextual semantics. In this paper we go beyond the vector representations at the word level and propose a novel framework that learns higher-level feature representations of n-grams, phrases and sentences using a deep neural network built from stacked Convolutional Restricted Boltzmann Machines (CRBMs). These representations have been shown to map syntactically and semantically related n-grams to closeby locations in the hidden feature space. We have experimented to additionally incorporate these higher-level features into supervised classifier training for two sentiment analysis tasks: subjectivity classification and sentiment classification. Our results have demonstrated the success of our proposed framework with 4% improvement in accuracy observed for subjectivity classification and improved the results achieved for sentiment classification over models trained without our higher level features.

Noisy fitness evaluation in genetic algorithms and the dynamics of learning

A theoretical model is presented which describes selection in a genetic algorithm (GA) under a stochastic fitness measure and correctly accounts for finite population effects. Although this model describes a number of selection schemes, we only consider Boltzmann selection in detail here as results for this form of selection are particularly transparent when fitness is corrupted by additive Gaussian noise. Finite population effects are shown to be of fundamental importance in this case, as the noise has no effect in the infinite population limit. In the limit of weak selection we show how the effects of any Gaussian noise can be removed by increasing the population size appropriately. The theory is tested on two closely related problems: the one-max problem corrupted by Gaussian noise and generalization in a perceptron with binary weights. The averaged dynamics can be accurately modelled for both problems using a formalism which describes the dynamics of the GA using methods from statistical mechanics. The second problem is a simple example of a learning problem and by considering this problem we show how the accurate characterization of noise in the fitness evaluation may be relevant in machine learning. The training error (negative fitness) is the number of misclassified training examples in a batch and can be considered as a noisy version of the generalization error if an independent batch is used for each evaluation. The noise is due to the finite batch size and in the limit of large problem size and weak selection we show how the effect of this noise can be removed by increasing the population size. This allows the optimal batch size to be determined, which minimizes computation time as well as the total number of training examples required.

The charge effect on the hindrance factors for diffusion and convection of a solute in pores II

The diffusion and convection of a solute suspended in a fluid across porous membranes are known to be reduced compared to those in a bulk solution, owing to the fluid mechanical interaction between the solute and the pore wall as well as steric restriction. If the solute and the pore wall are electrically charged, the electrostatic interaction between them could affect the hindrance to diffusion and convection. In this study, the transport of charged spherical solutes through charged circular cylindrical pores filled with an electrolyte solution containing small ions was studied numerically by using a fluid mechanical and electrostatic model. Based on a mean field theory, the electrostatic interaction energy between the solute and the pore wall was estimated from the Poisson-Boltzmann equation, and the charge effect on the solute transport was examined for the solute and pore wall of like charge. The results were compared with those obtained from the linearized form of the Poisson-Boltzmann equation, i.e.the Debye-Hückel equation. © 2012 The Japan Society of Fluid Mechanics and IOP Publishing Ltd.

Fluid mechanical and electrostatic study on the osmotic flow through cylindrical pores

When two solutions differing in solute concentration are separated by a porous membrane, the osmotic pressure will generate a net volume flux of the suspending fluid across the membrane; this is termed osmotic flow. We consider the osmotic flow across a membrane with circular cylindrical pores when the solute and the pore walls are electrically charged, and the suspending fluid is an electrolytic solution containing small cations and anions. Under the condition in which the radius of the pores and that of the solute molecules greatly exceed those of the solvent as well as the ions, a fluid mechanical and electrostatic theory is introduced to describe the osmotic flow in the presence of electric charge. The interaction energy, including the electrostatic interaction between the solute and the pore wall, plays a key role in determining the osmotic flow. We examine the electrostatic effect on the osmotic flow and discuss the difference in the interaction energy determined from the nonlinear Poisson-Boltzmann equation and from its linearized equation (the Debye-Hückel equation).

Electrical charge effect on osmotic flow through pores

An electrostatic model for osmotic flow through circular cylindrical pores is developed to describe the reflection coefficient for the membrane transport in the presence of surface charges on the pore wall and the solute. For a spherical solute placed at an arbitrary radial position in the pore, the electrical potential was computed by a spectral element method applied to the Poisson-Boltzmann equation together with the condition of electrical neutrality. The interaction energy between the surface charges was used to estimate the osmotic reflection coefficient. The proposed model predicts that even for a small Debye length compared to the pore radius, the repulsive electrostatic interaction between the surface charges could significantly increase the osmotic flow through the pore.

Assessment of physical and hydrological properties of biological soil crusts using x-ray microtomography and modelling

Biological soil crusts (BSCs) are formed by aggregates of soil particles and communities of microbial organisms and are common in all drylands. The role of BSCs on infiltration remains uncertain due to the lack of data on their role in affecting soil physical properties such as porosity and structure. Quantitative assessment of these properties is primarily hindered by the fragile nature of the crusts. Here we show how the use of a combination of non-destructive imaging X-ray microtomography (XMT) and Lattice Boltzmann method (LBM) enables quantification of key soil physical parameters and the modeling of water flow through BSCs samples from Kalahari Sands, Botswana. We quantify porosity and flow changes as a result of mechanical disturbance of such a fragile cyanobacteria-dominated crust. Results show significant variations in porosity between different types of crusts and how they affect the flow and that disturbance of a cyanobacteria-dominated crust results in the breakdown of larger pore spaces and reduces flow rates through the surface layer. We conclude that the XMT–LBM approach is well suited for study of fragile surface crust samples where physical and hydraulic properties cannot be easily quantified using conventional methods.

High-frequency conductivity of optically excited charge carriers in hydrogenated nanocrystalline silicon investigated by spectroscopic femtosecond pump-probe reflectivity measurements

We report an investigation into the high-frequency conductivity of optically excited charge carriers far from equilibrium with the lattice. The investigated samples consist of hydrogenated nanocrystalline silicon films grown on a thin film of silicon oxide on top of a silicon substrate. For the investigation, we used an optical femtosecond pump-probe setup to measure the reflectance change of a probe beam. The pump beam ranged between 580 and 820nm, whereas the probe wavelength spanned 770 to 810nm. The pump fluence was fixed at 0.6mJ/cm2. We show that at a fixed delay time of 300fs, the conductivity of the excited electron-hole plasma is described well by a classical conductivity model of a hot charge carrier gas found at Maxwell-Boltzmann distribution, while Fermi-Dirac statics is not suitable. This is corroborated by values retrieved from pump-probe reflectance measurements of the conductivity and its dependence on the excitation wavelength and carrier temperature. The conductivity decreases monotonically as a function of the excitation wavelength, as expected for a nondegenerate charge carrier gas.