Fast acoustic likelihood computation using low-rank matrix approximation

Acoustic modeling using mixtures of multivariate Gaussians is the prevalent approach for many speech processing problems. Computing likelihoods against a large set of Gaussians is required as a part of many speech processing systems and it is the computationally dominant phase for LVCSR systems. We express the likelihood computation as a multiplication of matrices representing augmented feature vectors and Gaussian parameters. The computational gain of this approach over traditional methods is by exploiting the structure of these matrices and efficient implementation of their multiplication.In particular, we explore direct low-rank approximation of the Gaussian parameter matrix and indirect derivation of low-rank factors of the Gaussian parameter matrix by optimum approximation of the likelihood matrix. We show that both the methods lead to similar speedups but the latter leads to far lesser impact on the recognition accuracy. Experiments on a 1138 word vocabulary RM1 task using Sphinx 3.7 system show that, for a typical case the matrix multiplication approach leads to overall speedup of 46%. Both the low-rank approximation methods increase the speedup to around 60%, with the former method increasing the word error rate (WER) from 3.2% to 6.6%, while the latter increases the WER from 3.2% to 3.5%.

A nested linear codes approach to distributed function computation over subspaces

In this paper, we consider a distributed function computation setting, where there are m distributed but correlated sources X1,...,Xm and a receiver interested in computing an s-dimensional subspace generated by [X1,...,Xm]Γ for some (m × s) matrix Γ of rank s. We construct a scheme based on nested linear codes and characterize the achievable rates obtained using the scheme. The proposed nested-linear-code approach performs at least as well as the Slepian-Wolf scheme in terms of sum-rate performance for all subspaces and source distributions. In addition, for a large class of distributions and subspaces, the scheme improves upon the Slepian-Wolf approach. The nested-linear-code scheme may be viewed as uniting under a common framework, both the Korner-Marton approach of using a common linear encoder as well as the Slepian-Wolf approach of employing different encoders at each source. Along the way, we prove an interesting and fundamental structural result on the nature of subspaces of an m-dimensional vector space V with respect to a normalized measure of entropy. Here, each element in V corresponds to a distinct linear combination of a set {Xi}im=1 of m random variables whose joint probability distribution function is given.

Linear Coding Schemes for the Distributed Computation of Subspaces

Let X-1,..., X-m be a set of m statistically dependent sources over the common alphabet F-q, that are linearly independent when considered as functions over the sample space. We consider a distributed function computation setting in which the receiver is interested in the lossless computation of the elements of an s-dimensional subspace W spanned by the elements of the row vector X-1,..., X-m]Gamma in which the (m x s) matrix Gamma has rank s. A sequence of three increasingly refined approaches is presented, all based on linear encoders. The first approach uses a common matrix to encode all the sources and a Korner-Marton like receiver to directly compute W. The second improves upon the first by showing that it is often more efficient to compute a carefully chosen superspace U of W. The superspace is identified by showing that the joint distribution of the {X-i} induces a unique decomposition of the set of all linear combinations of the {X-i}, into a chain of subspaces identified by a normalized measure of entropy. This subspace chain also suggests a third approach, one that employs nested codes. For any joint distribution of the {X-i} and any W, the sum-rate of the nested code approach is no larger than that under the Slepian-Wolf (SW) approach. Under the SW approach, W is computed by first recovering each of the {X-i}. For a large class of joint distributions and subspaces W, the nested code approach is shown to improve upon SW. Additionally, a class of source distributions and subspaces are identified, for which the nested-code approach is sum-rate optimal.

Runtime dependence computation and execution of loops on heterogeneous systems

GPUs have been used for parallel execution of DOALL loops. However, loops with indirect array references can potentially cause cross iteration dependences which are hard to detect using existing compilation techniques. Applications with such loops cannot easily use the GPU and hence do not benefit from the tremendous compute capabilities of GPUs. In this paper, we present an algorithm to compute at runtime the cross iteration dependences in such loops. The algorithm uses both the CPU and the GPU to compute the dependences. Specifically, it effectively uses the compute capabilities of the GPU to quickly collect the memory accesses performed by the iterations by executing the slice functions generated for the indirect array accesses. Using the dependence information, the loop iterations are levelized such that each level contains independent iterations which can be executed in parallel. Another interesting aspect of the proposed solution is that it pipelines the dependence computation of the future level with the actual computation of the current level to effectively utilize the resources available in the GPU. We use NVIDIA Tesla C2070 to evaluate our implementation using benchmarks from Polybench suite and some synthetic benchmarks. Our experiments show that the proposed technique can achieve an average speedup of 6.4x on loops with a reasonable number of cross iteration dependences.

An approximately H-1-optimal Petrov-Galerkin meshfree method: application to computation of scattered light for optical tomography

Nearly pollution-free solutions of the Helmholtz equation for k-values corresponding to visible light are demonstrated and verified through experimentally measured forward scattered intensity from an optical fiber. Numerically accurate solutions are, in particular, obtained through a novel reformulation of the H-1 optimal Petrov-Galerkin weak form of the Helmholtz equation. Specifically, within a globally smooth polynomial reproducing framework, the compact and smooth test functions are so designed that their normal derivatives are zero everywhere on the local boundaries of their compact supports. This circumvents the need for a priori knowledge of the true solution on the support boundary and relieves the weak form of any jump boundary terms. For numerical demonstration of the above formulation, we used a multimode optical fiber in an index matching liquid as the object. The scattered intensity and its normal derivative are computed from the scattered field obtained by solving the Helmholtz equation, using the new formulation and the conventional finite element method. By comparing the results with the experimentally measured scattered intensity, the stability of the solution through the new formulation is demonstrated and its closeness to the experimental measurements verified.

Nearly Constant Switching Frequency Hysteresis Current Controller with Fast Online Computation of Boundary for a 2-Level Induction Motor Drive

A nearly constant switching frequency current hysteresis controller for a 2-level inverter fed induction motor drive is proposed in this paper: The salient features of this controller are fast dynamics for the current, inherent protection against overloads and less switching frequency variation. The large variation of switching frequency as in the conventional hysteresis controller is avoided by defining a current-error boundary which is obtained from the current-error trajectory of the standard space vector PWM. The current-error boundary is computed at every sampling interval based on the induction machine parameters and from the estimated fundamental stator voltage. The stator currents are always monitored and when the current-error exceeds the boundary, voltage space vector is switched to reduce the current-error. The proposed boundary computation algorithm is applicable in linear and over-modulation region and it is simple to implement in any standard digital signal processor: Detailed experimental verification is done using a 7.5 kW induction motor and the results are given to show the performance of the drive at various operating conditions and validate the proposed advantages.

Hysteresis current controller for a general n-level inverter fed drive with online current error boundary computation and nearly constant switching frequency

A space vector-based hysteresis current controller for any general n-level three phase inverter fed induction motor drive is proposed in this study. It offers fast dynamics, inherent overload protection and low harmonic distortion for the phase voltages and currents. The controller performs online current error boundary calculations and a nearly constant switching frequency is obtained throughout the linear modulation range. The proposed scheme uses only the adjacent voltage vectors of the present sector, similar to space vector pulse-width modulation and exhibits fast dynamic behaviour under different transient conditions. The steps involved in the boundary calculation include the estimation of phase voltages from the current ripple, computation of switching time and voltage error vectors. Experimental results are given to show the performance of the drive at various speeds, effect of sudden change of the load, acceleration, speed reversal and validate the proposed advantages.

Faster computation of the Karhunen-Loeve expansion using its domain independence property

The goal of this work is to reduce the cost of computing the coefficients in the Karhunen-Loeve (KL) expansion. The KL expansion serves as a useful and efficient tool for discretizing second-order stochastic processes with known covariance function. Its applications in engineering mechanics include discretizing random field models for elastic moduli, fluid properties, and structural response. The main computational cost of finding the coefficients of this expansion arises from numerically solving an integral eigenvalue problem with the covariance function as the integration kernel. Mathematically this is a homogeneous Fredholm equation of second type. One widely used method for solving this integral eigenvalue problem is to use finite element (FE) bases for discretizing the eigenfunctions, followed by a Galerkin projection. This method is computationally expensive. In the current work it is first shown that the shape of the physical domain in a random field does not affect the realizations of the field estimated using KL expansion, although the individual KL terms are affected. Based on this domain independence property, a numerical integration based scheme accompanied by a modification of the domain, is proposed. In addition to presenting mathematical arguments to establish the domain independence, numerical studies are also conducted to demonstrate and test the proposed method. Numerically it is demonstrated that compared to the Galerkin method the computational speed gain in the proposed method is of three to four orders of magnitude for a two dimensional example, and of one to two orders of magnitude for a three dimensional example, while retaining the same level of accuracy. It is also shown that for separable covariance kernels a further cost reduction of three to four orders of magnitude can be achieved. Both normal and lognormal fields are considered in the numerical studies. (c) 2014 Elsevier B.V. All rights reserved.

Quick, Decentralized, Energy-Efficient One-Shot Max Function Computation Using Timer-Based Selection

In several wireless sensor networks, it is of interest to determine the maximum of the sensor readings and identify the sensor responsible for it. We propose a novel, decentralized, scalable, energy-efficient, timer-based, one-shot max function computation (TMC) algorithm. In it, the sensor nodes do not transmit their readings in a centrally pre-defined sequence. Instead, the nodes are grouped into clusters, and computation occurs over two contention stages. First, the nodes in each cluster contend with each other using the timer scheme to transmit their reading to their cluster-heads. Thereafter, the cluster-heads use the timer scheme to transmit the highest sensor reading in their cluster to the fusion node. One new challenge is that the use of the timer scheme leads to collisions, which can make the algorithm fail. We optimize the algorithm to minimize the average time required to determine the maximum subject to a constraint on the probability that it fails to find the maximum. TMC significantly lowers average function computation time, average number of transmissions, and average energy consumption compared to approaches proposed in the literature.

Computation of power released during corona treatment on polymeric insulators under ac and dc excitation

With increasing energy demand, it necessitates to generate and transmit the electrical power with minimal losses. High voltage power transmission is the most economical way of transmitting bulk power over long distances. Transmission insulator is one of the main components used as a mechanical support and to electrically isolate the conductor from the tower. Corona from the hardware and conductors can significantly affect the performance of the polymeric insulators. In the present investigation a methodology is presented to evaluate the corona performance of the polymeric shed material under different environment conditions for both ac and dc excitation. The results of the comprehensive analysis on various polymeric samples and the power released from the corona electrode for both the ac and dc excitation are presented. Some interesting results obtained from the chemical analysis confirmed the presence of nitric acid species on the treated sample which in long term will affect the strength of the insulator, also the morphological changes were found to be varying for different experimental conditions. (C) 2015 The Authors. Published by Elsevier Ltd.

Comparison of Matching Pursuit Algorithm with Other Signal Processing Techniques for Computation of the Time-Frequency Power Spectrum of Brain Signals

Signals recorded from the brain often show rhythmic patterns at different frequencies, which are tightly coupled to the external stimuli as well as the internal state of the subject. In addition, these signals have very transient structures related to spiking or sudden onset of a stimulus, which have durations not exceeding tens of milliseconds. Further, brain signals are highly nonstationary because both behavioral state and external stimuli can change on a short time scale. It is therefore essential to study brain signals using techniques that can represent both rhythmic and transient components of the signal, something not always possible using standard signal processing techniques such as short time fourier transform, multitaper method, wavelet transform, or Hilbert transform. In this review, we describe a multiscale decomposition technique based on an over-complete dictionary called matching pursuit (MP), and show that it is able to capture both a sharp stimulus-onset transient and a sustained gamma rhythm in local field potential recorded from the primary visual cortex. We compare the performance of MP with other techniques and discuss its advantages and limitations. Data and codes for generating all time-frequency power spectra are provided.

Computation of Viscous Incompressible Flow using Pressure Correction Method on Unstructured Chimera Grid

In this paper, a pressure correction algorithm for computing incompressible flows is modified and implemented on unstructured Chimera grid. Schwarz method is used to couple the solutions of different sub-domains. A new interpolation to ensure consistency between primary variables and auxiliary variables is proposed. Other important issues such as global mass conservation and order of accuracy in the interpolations are also discussed. Two numerical simulations are successfully performed. They include one steady case, the lid-driven cavity and one unsteady case, the flow around a circular cylinder. The results demonstrate a very good performance of the proposed scheme on unstructured Chimera grids. It prevents the decoupling of pressure field in the overlapping region and requires only little modification to the existing unstructured Navier–Stokes (NS) solver. The numerical experiments show the reliability and potential of this method in applying to practical problems.