A variational perspective on noise-robust speech recognition

Model compensation methods for noise-robust speech recognition have shown good performance. Predictive linear transformations can approximate these methods to balance computational complexity and compensation accuracy. This paper examines both of these approaches from a variational perspective. Using a matched-pair approximation at the component level yields a number of standard forms of model compensation and predictive linear transformations. However, a tighter bound can be obtained by using variational approximations at the state level. Both model-based and predictive linear transform schemes can be implemented in this framework. Preliminary results show that the tighter bound obtained from the state-level variational approach can yield improved performance over standard schemes. © 2011 IEEE.

Two problems with variational expectation maximisation for time-series models

Variational methods are a key component of the approximate inference and learning toolbox. These methods fill an important middle ground, retaining distributional information about uncertainty in latent variables, unlike maximum a posteriori methods (MAP), and yet generally requiring less computational time than Monte Carlo Markov Chain methods. In particular the variational Expectation Maximisation (vEM) and variational Bayes algorithms, both involving variational optimisation of a free-energy, are widely used in time-series modelling. Here, we investigate the success of vEM in simple probabilistic time-series models. First we consider the inference step of vEM, and show that a consequence of the well-known compactness property of variational inference is a failure to propagate uncertainty in time, thus limiting the usefulness of the retained distributional information. In particular, the uncertainty may appear to be smallest precisely when the approximation is poorest. Second, we consider parameter learning and analytically reveal systematic biases in the parameters found by vEM. Surprisingly, simpler variational approximations (such a mean-field) can lead to less bias than more complicated structured approximations.

Research on the influence of fabrication parameters on the performance of buried ion-exchanged glass waveguides using the variational method

The variational method is proposed to analyze the influence of the fabrication parameters on the performance of buried K+-Na+ ion-exchanged Er3+-Yb3+ ions co-doped glass waveguide. The unknown parameters of the Hermite-Gaussian functions as the trial field distribution are determined based on the scalar variational principle. It is demonstrated that the results calculated in this paper agree with those measured in the experiment. The mode dimensions, the effective refractive index, and the overlap factor as the functions of the fabrication parameters are investigated. These results of the variational analysis are useful for the design and optimization of Er3+-Yb3+ ions co-doped waveguides.

Variational calculations of neutral bound excitons in GaAs quantum-well wires

The binding energy of an exciton bound to a neutral donor (D-0,X) in GaAs quantum-well wires is calculated variationally as a function of the wire width for different positions of the impurity inside the wire by using a two-parameter wavefunction. There is no artificial parameter added in our calculation. The results we have obtained show that the binding energies are closely correlated to the sizes of the wire, the impurity position, and also that their magnitudes are greater than those in the two-dimensional quantum wells compared. In addition, we also calculate the average interparticle distance as a function of the wire width. The results are discussed in detail.

Variational calculations of ionized-donor-bound excitons in GaAs-AlxGa1-xAs quantum wells

Using a two-parameter wave function, we calculate variationally the binding energy of an exciton bound to an ionized donor impurity (D+,X) in GaAs-AlxGa1-xAs quantum wells for the values of the well width from 10 to 300 Angstrom, when the dopant is located in the center of the well and at the edge of the well. The theoretical results confirm that the previous experimental speculation proposed by Reynolds tit al. [Phys. Rev. B 40, 6210 (1989)] is the binding energy of D+,X for the dopant at the edge of the well. in addition, we also calculate the center-of-mass wave function of the exciton and the average interparticle distances. The results are discussed in detail.

An eigenfunction expansion-variational method in prediction of the transverse thermal conductivity of fiber reinforced composites considering interfacial characteristics

An eigenfunction expansion-variational method based on a unit cell is developed to deal with the steady-state heat conduction problem of doubly-periodic fiber reinforced composites with interfacial thermal contact resistance or coating. The numerical results show a rapid convergence of the present method. The present solution provides a unified first-order approximation formula of the effective thermal conductivity for different interfacial characteristics and fiber distributions. A comparison with the present high-order results, available experimental data and micromechanical estimations demonstrates that the first-order approximation formula is a good engineering closed-form formula. An engineering equivalent parameter reflecting the overall influence of the thermal conductivities of the matrix and fibers and the interfacial characteristic on the effective thermal conductivity, is found. The equivalent parameter can greatly simplify the complicated relation of the effective thermal conductivity to the internal structure of a composite. (c) 2010 Elsevier Ltd. All rights reserved.