The infinite HMM for unsupervised PoS tagging

We extend previous work on fully unsupervised part-of-speech tagging. Using a non-parametric version of the HMM, called the infinite HMM (iHMM), we address the problem of choosing the number of hidden states in unsupervised Markov models for PoS tagging. We experiment with two non-parametric priors, the Dirichlet and Pitman-Yor processes, on the Wall Street Journal dataset using a parallelized implementation of an iHMM inference algorithm. We evaluate the results with a variety of clustering evaluation metrics and achieve equivalent or better performances than previously reported. Building on this promising result we evaluate the output of the unsupervised PoS tagger as a direct replacement for the output of a fully supervised PoS tagger for the task of shallow parsing and compare the two evaluations. © 2009 ACL and AFNLP.

An Infinite Latent Attribute Model for Network Data

Latent variable models for network data extract a summary of the relational structure underlying an observed network. The simplest possible models subdivide nodes of the network into clusters; the probability of a link between any two nodes then depends only on their cluster assignment. Currently available models can be classified by whether clusters are disjoint or are allowed to overlap. These models can explain a "flat" clustering structure. Hierarchical Bayesian models provide a natural approach to capture more complex dependencies. We propose a model in which objects are characterised by a latent feature vector. Each feature is itself partitioned into disjoint groups (subclusters), corresponding to a second layer of hierarchy. In experimental comparisons, the model achieves significantly improved predictive performance on social and biological link prediction tasks. The results indicate that models with a single layer hierarchy over-simplify real networks.

Investigation of the dilution process for measurement of particulate matter from spark-ignition engines

Measurements of particulate matter (PM) from spark ignition (SI) engine exhaust using dilution tunnels will become more prevalent as emission standards are tightened. Hence, a study of the dilution process was undertaken in order to understand how various dilution related parameters affect the accuracy with which PM sizes and concentrations can be determined. A SI and a compression ignition (CI) engine were separately used to examine parameters of the dilution process; the present work discusses the results in the context of SI exhaust dilution. A Scanning Mobility Particle Sizer (SMPS) was used to measure the size distribution, number density, and volume fraction of PM. Temperature measurements in the exhaust pipe and dilution tunnel reveal the degree of mixing between exhaust and dilution air, the effect of flowrate on heat transfer from undiluted and diluted exhaust to the environment, and the minimum permissible dilution ratio for a maximum sample temperature of 52°C. Measurements of PM concentrations as a function of dilution ratio show the competing effects of temperature and particle/vapor concentrations on particle growth dynamics, which result in a range of dilution ratios-from 13 to 18-where the effect of dilution ratio, independent of flowrate, is kept to a minimum. This range of dilution ratios is therefore optimal in order to achieve repeatable PM concentration measurements. Particle dynamics during transit through the tunnel operating at the optimal dilution ratio was found statistically insignificant compared to data scatter. Such small differences in number concentration may be qualitatively representative of particle losses for SI exhaust, but small increases in PM volume fraction during transit through the tunnel may significantly underestimate accretion of mass due to unburned hydrocarbons (HCs) emitted by SI engines. The fraction of SI-derived PM mass due to adsorbed/absorbed vapor, estimated from these data, is consistent with previous chemical analyses of PM. © 1998 Society of Automotive Engineers, Inc.

Local stability results for the collective behaviors of infinite populations of pulse-coupled oscillators

In this paper, we investigate the behavior of pulse-coupled integrate-and-fire oscillators. Because the stability analysis of finite populations is intricate, we investigate stability results in the approximation of infinite populations. In addition to recovering known stability results of finite populations, we also obtain new stability results for infinite populations. In particular, under a weak coupling assumption, we solve for the continuum model a conjecture still prevailing in the finite dimensional case. © 2011 IEEE.

The Supervised IBP: Neighbourhood Preserving Infinite Latent Feature Models

We propose a probabilistic model to infer supervised latent variables in the Hamming space from observed data. Our model allows simultaneous inference of the number of binary latent variables, and their values. The latent variables preserve neighbourhood structure of the data in a sense that objects in the same semantic concept have similar latent values, and objects in different concepts have dissimilar latent values. We formulate the supervised infinite latent variable problem based on an intuitive principle of pulling objects together if they are of the same type, and pushing them apart if they are not. We then combine this principle with a flexible Indian Buffet Process prior on the latent variables. We show that the inferred supervised latent variables can be directly used to perform a nearest neighbour search for the purpose of retrieval. We introduce a new application of dynamically extending hash codes, and show how to effectively couple the structure of the hash codes with continuously growing structure of the neighbourhood preserving infinite latent feature space.

A reversible infinite HMM using normalised random measures

Copyright 2014 by the author(s). We present a nonparametric prior over reversible Markov chains. We use completely random measures, specifically gamma processes, to construct a countably infinite graph with weighted edges. By enforcing symmetry to make the edges undirected we define a prior over random walks on graphs that results in a reversible Markov chain. The resulting prior over infinite transition matrices is closely related to the hierarchical Dirichlet process but enforces reversibility. A reinforcement scheme has recently been proposed with similar properties, but the de Finetti measure is not well characterised. We take the alternative approach of explicitly constructing the mixing measure, which allows more straightforward and efficient inference at the cost of no longer having a closed form predictive distribution. We use our process to construct a reversible infinite HMM which we apply to two real datasets, one from epigenomics and one ion channel recording.

Wave dispersion and power flow analysis of an infinite aerofoil-shaped beam

Classic flutter analysis models an aerofoil as a two degree-of-freedom rigid body supported by linear and torsional springs, which represent the bending and torsional stiffness of the aerofoil section. In this classic flutter model, no energy transfer or dissipation can occur in the span-wise direction of the aerofoil section. However, as the aspect ratio of an aerofoil section increases, this span-wise energy transfer - in the form of travelling waves - becomes important to the overall system dynamics. This paper extends the classic flutter model to include travelling waves in the span-wise direction. Namely, wave dispersion and power flow analysis of an infinite, aerofoil-shaped beam, subject to bending, torsion, tension and a constant wind excitation, is used to investigate the overall system stability. Examples of potential applications for these high aspect ratio aerofoil sections include high-altitude balloon tethers, towed cables, offshore risers and mooring lines.

Measurement of 5-day biochemical oxygen demand without sample dilution or bacterial and nutrient enhancement

A direct method for measuring the 5-day biochemical oxygen demand (BODS) of aquaculture samples that does not require sample dilution or bacterial and nutrient enrichment was evaluated. The regression coefficient (R-2) between the direct method and the standard method for the analyses of 32 samples from catfish ponds was 0.996. The slope of the regression line did not differ from 1.0 or the Y-intercept from 0.0 at P = 0.05. Thus, there was almost perfect agreement between the two methods. The control limits (three standard deviations of the mean) for a standard solution containing 15 mg/L each of glutamic acid and glucose were 17.4 and 20.4 mg/L. The precision of the two methods, based on eight replicate analyses of four pond water samples did not differ at P = 0.05. (c) 2005 Elsevier B.V All rights reserved.

Impacts of hydrogen dilution on growth and optical properties of a-SiC : H films

Hydrogenated amorphous silicon-carbon (a-SiC:H) films were deposited by plasma enhanced chemical vapor deposition (PECVD) with a fixed methane to silane ratio ([CH4]/[SiH4]) of 1.2 and a wide range of hydrogen dilution (R-H=[H-2]/[SiH4 + CH4]) values of 12, 22, 33, 102 and 135. The impacts of RH on the structural and optical properties of the films were investigated by using UV-VIS transmission, Fourier transform infrared (FTIR) absorption, Raman scattering and photoluminescence (PL) measurements. The effects of high temperature annealing on the films were also probed. It is found that with increasing hydrogen dilution, the optical band gap increases, and the PL peak blueshifts from similar to1.43 to 1.62 eV. In annealed state, the room temperature PL peak for the low R-H samples disappears, while the PL peak for the high R-H samples appears at similar to 2.08 eV, which is attributed to nanocrystalline Si particles confined by Si-C and Si-O bonds.

The compact microcrystalline Si thin film with structure uniformity in the growth direction by hydrogen dilution profile

The hydrogen dilution profiling (HDP) technique has been developed to improve the quality and the crystalline uniformity in the growth direction of mu c-Si:H thin films prepared by hot-wire chemical-vapor deposition. The high H dilution in the initial growth stage reduces the amorphous transition layer from 30-50 to less than 10 nm. The uniformity of crystalline content X-c in the growth direction was much improved by the proper design of hydrogen dilution profiling which effectively controls the nonuniform transition region of Xc from 300 to less than 30 nm. Furthermore, the HDP approach restrains the formation of microvoids in mu c-Si: H thin films with a high Xc and enhances the compactness of the film. As a result the stability of mu c-Si: H thin films by HDP against the oxygen diffusion, as well as the electrical property, is much improved. (c) 2005 American Institute of Physics.

Impact of hydrogen dilution on microstructure and optoelectronic properties of silicon films deposited using tirisilane

We explored the deposition of hydrogenated amorphous silicon (a-Si: H) using trisilane (Si3H8) as a gas precursor in a radiofrequency plasma enhanced chemical vapour deposition process and studied the suitability of this material for photovoltaic applications. The impact of hydrogen dilution on the deposition rate and microstructure of the films is systematically examined. Materials deposited using trisilane are compared with that using disilane (Si2H6). It is found that when using Si3H8 as the gas precursor the deposition rate increases by a factor of similar to 1.5 for the same hydrogen dilution (R = [H-2]/[Si3H8] or [H-2]/[Si2H6])- Moreover, the structural transition from amorphous to nanocrystalline occurs at a higher hydrogen dilution level for Si3H8 and the transition is more gradual as compared with Si2H6 deposited films. Single-junction n-i-p a-Si: H solar cells were prepared with intrinsic layers deposited using Si3H8 or Si2H6. The dependence of open circuit voltage (V-oc) on hydrogen dilution was investigated. V-oc greater than 1 V can be obtained when the i-layers are deposited at a hydrogen dilution of 180 and 100 using Si3H8 and Si2H6, respectively.

Spontaneous emission lifetime distribution of infinite line antennas in two-dimensional photonic crystals with finite size

The authors calculate the lifetime distribution functions of spontaneous emission from infinite line antennas embedded in two-dimensional disordered photonic crystals with finite size. The calculations indicate the coexistence of both accelerated and inhibited decay processes in disordered photonic crystals with finite size. The decay behavior of the spontaneous emission from infinite line antennas changes significantly by varying factors such as the line antennas' positions in the disordered photonic crystal, the shape of the crystal, the filling fraction, and the dielectric constant. Moreover, the authors analyze the effect of the degree of disorder on spontaneous emission. (c) 2007 American Institute of Physics.

High quality microcrystalline Si films by hydrogen dilution profile

Novel hydrogen dilution profiling (HDP) technique was developed to improve the uniformity in the growth direction of mu c-Si:H thin films prepared by hot wire chemical vapor deposition (HWCVD). It was found that the high H dilution ratio reduces the incubation layer from 30 nm to less than 10 nm. A proper design of hydrogen dilution profiling improves the uniformity of crystalline content, X-c, in the growth direction and restrains the formation of micro-voids as well. As a result the compactness of mu c-Si:H films with a high crystalline content is enhanced and the stability of mu c-Si:H thin film against the oxygen diffusion is much improved. Meanwhile the HDP mu c-Si:H films exhibit the low defect states. The high nucleation density from high hydrogen dilution at early stage is a critical parameter to improve the quality of mu c-Si:H films. (c) 2006 Published by Elsevier B.V.