Exponentiated gradient algorithms for conditional random fields and max-margin Markov Networks

Log-linear and maximum-margin models are two commonly-used methods in supervised machine learning, and are frequently used in structured prediction problems. Efficient learning of parameters in these models is therefore an important problem, and becomes a key factor when learning from very large data sets. This paper describes exponentiated gradient (EG) algorithms for training such models, where EG updates are applied to the convex dual of either the log-linear or max-margin objective function; the dual in both the log-linear and max-margin cases corresponds to minimizing a convex function with simplex constraints. We study both batch and online variants of the algorithm, and provide rates of convergence for both cases. In the max-margin case, O(1/ε) EG updates are required to reach a given accuracy ε in the dual; in contrast, for log-linear models only O(log(1/ε)) updates are required. For both the max-margin and log-linear cases, our bounds suggest that the online EG algorithm requires a factor of n less computation to reach a desired accuracy than the batch EG algorithm, where n is the number of training examples. Our experiments confirm that the online algorithms are much faster than the batch algorithms in practice. We describe how the EG updates factor in a convenient way for structured prediction problems, allowing the algorithms to be efficiently applied to problems such as sequence learning or natural language parsing. We perform extensive evaluation of the algorithms, comparing them to L-BFGS and stochastic gradient descent for log-linear models, and to SVM-Struct for max-margin models. The algorithms are applied to a multi-class problem as well as to a more complex large-scale parsing task. In all these settings, the EG algorithms presented here outperform the other methods.

Mean and variance of estimates of the bispectrum of a harmonic random process-an analysis including leakage effects

Analytical expressions are derived for the mean and variance, of estimates of the bispectrum of a real-time series assuming a cosinusoidal model. The effects of spectral leakage, inherent in discrete Fourier transform operation when the modes present in the signal have a nonintegral number of wavelengths in the record, are included in the analysis. A single phase-coupled triad of modes can cause the bispectrum to have a nonzero mean value over the entire region of computation owing to leakage. The variance of bispectral estimates in the presence of leakage has contributions from individual modes and from triads of phase-coupled modes. Time-domain windowing reduces the leakage. The theoretical expressions for the mean and variance of bispectral estimates are derived in terms of a function dependent on an arbitrary symmetric time-domain window applied to the record. the number of data, and the statistics of the phase coupling among triads of modes. The theoretical results are verified by numerical simulations for simple test cases and applied to laboratory data to examine phase coupling in a hypothesis testing framework

Compilation of somatic mutations of the CDKN2 gene in human cancers : non-random distribution of base substitutions

The CDKN2 gene, encoding the cyclin-dependent kinase inhibitor p16, is a tumour suppressor gene that maps to chromosome band 9p21-p22. The most common mechanism of inactivation of this gene in human cancers is through homozygous deletion; however, in a smaller proportion of tumours and tumour cell lines intragenic mutations occur. In this study we have compiled a database of over 120 published point mutations in the CDKN2 gene from a wide variety of tumour types. A further 50 deletions, insertions, and splice mutations in CDKN2 have also been compiled. Furthermore, we have standardised the numbering of all mutations according to the full-length 156 amino acid form of p16. From this study we are able to define several hot spots, some of which occur at conserved residues within the ankyrin domains of p16. While many of the hotspots are shared by a number of cancers, the relative importance of each position varies, possibly reflecting the role of different carcinogens in the development of certain tumours. As reported previously, the mutational spectrum of CDKN2 in melanomas differs from that of internal malignancies and supports the involvement of UV in melanoma tumorigenesis. Notably, 52% of all substitutions in melanoma-derived samples occurred at just six nucleotide positions. Nonsense mutations comprise a comparatively high proportion of mutations present in the CDKN2 gene, and possible explanations for this are discussed.

Innovative planning approaches for generator interconnection group study

Today, a large number of wind generator interconnection requests have been queued and are being processed. The generator interconnection group study is a way to reduce the generator interconnection cycle time and increase interconnection certainty. However, it is very challenging to identify the “best” transmission upgrades for a large group of generator interconnections. It is also very important to differentiate the constraints caused by each generator interconnection request and identify their responsibilities for transmission upgrades. This paper outlines some innovative study approaches that can be used in a group study with large numbers of generator interconnection requests in a constrained area. Improved study methods are introduced, and a summary and conclusions are derived from the study.

On the probability density function of the product of Rayleigh distributed random variables

In this paper we investigate the distribution of the product of Rayleigh distributed random variables. Considering the Mellin-Barnes inversion formula and using the saddle point approach we obtain an upper bound for the product distribution. The accuracy of this tail-approximation increases as the number of random variables in the product increase.

Generator coherency and area detection in large power systems

This paper provides a new general approach for defining coherent generators in power systems based on the coherency in low frequency inter-area modes. The disturbance is considered to be distributed in the network by applying random load changes which is the random walk representation of real loads instead of a single fault and coherent generators are obtained by spectrum analysis of the generators velocity variations. In order to find the coherent areas and their borders in the inter-connected networks, non-generating buses are assigned to each group of coherent generator using similar coherency detection techniques. The method is evaluated on two test systems and coherent generators and areas are obtained for different operating points to provide a more accurate grouping approach which is valid across a range of realistic operating points of the system.

Random Breath Testing in Australia: Is there an Optimum Level of Intensity?

Background: Random Breath Testing (RBT) is the main drink driving law enforcement tool used throughout Australia. International comparative research considers Australia to have the most successful RBT program compared to other countries in terms of crash reductions (Erke, Goldenbeld, & Vaa, 2009). This success is attributed to the programs high intensity (Erke et al., 2009). Our review of the extant literature suggests that there is no research evidence that indicates an optimal level of alcohol breath testing. That is, we suggest that no research exists to guide policy regarding whether or not there is a point at which alcohol related crashes reach a point of diminishing returns as a result of either saturated or targeted RBT testing. Aims: In this paper we first provide an examination of RBTs and alcohol related crashes across Australian jurisdictions. We then address the question of whether or not an optimal level of random breath testing exists by examining the relationship between the number of RBTs conducted and the occurrence of alcohol-related crashes over time, across all Australian states. Method: To examine the association between RBT rates and alcohol related crashes and to assess whether an optimal ratio of RBT tests per licenced drivers can be determined we draw on three administrative data sources form each jurisdiction. Where possible data collected spans January 1st 2000 to September 30th 2012. The RBT administrative dataset includes the number of Random Breath Tests (RBTs) conducted per month. The traffic crash administrative dataset contains aggregated monthly count of the number of traffic crashes where an individual’s recorded BAC reaches or exceeds 0.05g/ml of alcohol in blood. The licenced driver data were the monthly number of registered licenced drivers spanning January 2000 to December 2011. Results: The data highlights that the Australian story does not reflective of all States and territories. The stable RBT to licenced driver ratio in Queensland (of 1:1) suggests a stable rate of alcohol related crash data of 5.5 per 100,000 licenced drivers. Yet, in South Australia were a relative stable rate of RBT to licenced driver ratio of 1:2 is maintained the rate of alcohol related traffic crashes is substantially less at 3.7 per 100,000. We use joinpoint regression techniques and varying regression models to fit the data and compare the different patterns between jurisdictions. Discussion: The results of this study provide an updated review and evaluation of RBTs conducted in Australia and examines the association between RBTs and alcohol related traffic crashes. We also present an evidence base to guide policy decisions for RBT operations.

Are more city dwellers caught drink driving than country folk? An analysis by random breath testing apprehension rates

Background Random Breath Testing (RBT) remains a central enforcement strategy to deter and apprehend drink drivers in Queensland (Australia). Despite this, there is little published research regarding the exact drink driving apprehension rates across the state as measured through RBT activities. Aims The aim of the current study was to examine the prevalence of apprehending drink drivers in urban versus rural areas. Methods The Queensland Police Service provided data relating to the number of RBT conducted and apprehensions for the period 1 January 2000 to 31 December 2011. Results In the period, 35,082,386 random breath tests (both mobile and stationary) were conducted in Queensland which resulted in 248,173 individuals being apprehended for drink driving offences. Overall drink driving apprehension rates appear to have decreased across time. Close examination of the data revealed that the highest proportion of drink driving apprehensions (when compared with RBT testing rates) was in the Northern and Far Northern regions of Queensland (e.g., rural areas). In contrast, the lowest proportions were observed within the two Brisbane metropolitan regions (e.g., urban areas). However, differences in enforcement styles across the urban and rural regions need to be considered. Discussion and conclusions The research presentation will further outline the major findings of the study in regards to maximising the efficiency of RBT operations both within urban and rural areas of Queensland, Australia.

Spatio-temporal modelling of ultrafine particle number concentration

This thesis developed semi-parametric regression models for estimating the spatio-temporal distribution of outdoor airborne ultrafine particle number concentration (PNC). The models developed incorporate multivariate penalised splines and random walks and autoregressive errors in order to estimate non-linear functions of space, time and other covariates. The models were applied to data from the "Ultrafine Particles from Traffic Emissions and Child" project in Brisbane, Australia, and to longitudinal measurements of air quality in Helsinki, Finland. The spline and random walk aspects of the models reveal how the daily trend in PNC changes over the year in Helsinki and the similarities and differences in the daily and weekly trends across multiple primary schools in Brisbane. Midday peaks in PNC in Brisbane locations are attributed to new particle formation events at the Port of Brisbane and Brisbane Airport.

Random breath testing in Queensland and Western Australia : Examination of how the random breath testing rate influences alcohol related traffic crash rates

In this paper we explore the relationship between monthly random breath testing (RBT) rates (per 1000 licensed drivers) and alcohol-related traffic crash (ARTC) rates over time, across two Australian states: Queensland and Western Australia. We analyse the RBT, ARTC and licensed driver rates across 12 years; however, due to administrative restrictions, we model ARTC rates against RBT rates for the period July 2004 to June 2009. The Queensland data reveals that the monthly ARTC rate is almost flat over the five year period. Based on the results of the analysis, an average of 5.5 ARTCs per 100,000 licensed drivers are observed across the study period. For the same period, the monthly rate of RBTs per 1000 licensed drivers is observed to be decreasing across the study with the results of the analysis revealing no significant variations in the data. The comparison between Western Australia and Queensland shows that Queensland's ARTC monthly percent change (MPC) is 0.014 compared to the MPC of 0.47 for Western Australia. While Queensland maintains a relatively flat ARTC rate, the ARTC rate in Western Australia is increasing. Our analysis reveals an inverse relationship between ARTC RBT rates, that for every 10% increase in the percentage of RBTs to licensed driver there is a 0.15 decrease in the rate of ARTCs per 100,000 licenced drivers. Moreover, in Western Australia, if the 2011 ratio of 1:2 (RBTs to annual number of licensed drivers) were to double to a ratio of 1:1, we estimate the number of monthly ARTCs would reduce by approximately 15. Based on these findings we believe that as the number of RBTs conducted increases the number of drivers willing to risk being detected for drinking driving decreases, because the perceived risk of being detected is considered greater. This is turn results in the number of ARTCs diminishing. The results of this study provide an important evidence base for policy decisions for RBT operations.

Doubly Fed Induction Generator for wind energy generation using nine-switch power converter

A Three-Phase Nine-Switch Converter (NSC) topology for Doubly Fed Induction Generator in wind energy generation is proposed in this paper. This converter topology was used in various applications such as Hybrid Electric Vehicles and Uninterruptable Power Supplies. In this paper, Nine-Switch Converter is introduced in Doubly Fed Induction Generator in renewable energy application for the first time. It replaces the conventional Back-to-Back Pulse Width Modulated voltage source converter (VSC) which composed of twelve switches in many DFIG applications. Reduction in number of switches is the most beneficial in terms of cost and power switching losses. The operation principle of Nine-Switch Converter using SPWM method is discussed. The resulting NSC performance of rotor side current control, active power and reactive control are compared with Back-to Back voltage source converter performance. DC link voltage regulation using front end converter is also presented. Finally the simulation results of DFIG performances using NSC and Back-to-Back VSC are analyzed and compared.

Factors influencing robustness and effectiveness of conditional random fields in active learning frameworks

Active learning approaches reduce the annotation cost required by traditional supervised approaches to reach the same effectiveness by actively selecting informative instances during the learning phase. However, effectiveness and robustness of the learnt models are influenced by a number of factors. In this paper we investigate the factors that affect the effectiveness, more specifically in terms of stability and robustness, of active learning models built using conditional random fields (CRFs) for information extraction applications. Stability, defined as a small variation of performance when small variation of the training data or a small variation of the parameters occur, is a major issue for machine learning models, but even more so in the active learning framework which aims to minimise the amount of training data required. The factors we investigate are a) the choice of incremental vs. standard active learning, b) the feature set used as a representation of the text (i.e., morphological features, syntactic features, or semantic features) and c) Gaussian prior variance as one of the important CRFs parameters. Our empirical findings show that incremental learning and the Gaussian prior variance lead to more stable and robust models across iterations. Our study also demonstrates that orthographical, morphological and contextual features as a group of basic features play an important role in learning effective models across all iterations.

Experimental studies on comparison of microwave curing and thermal curing of epoxy resins used for alternative mould materials

In this present work attempts have been made to study the glass transition temperature of alternative mould materials by using both microwave heating and conventional oven heating. In this present work three epoxy resins, namely R2512, R2515 and R2516, which are commonly used for making injection moulds have been used in combination with two hardeners H2403 and H2409. The magnetron microwave generator used in this research is operating at a frequency of 2.45 GHz with a hollow rectangular waveguide. In order to distinguish the effects between the microwave and conventional heating, a number of experiments were performed to test their mechanical properties such as tensile and flexural strengths. Additionally, differential scanning calorimeter technique was implemented to measure the glass transition temperature on both microwave and conventional heating. This study provided necessary evidences to establish that microwave heated mould materials resulted with higher glass transition temperature than the conventional heating. Finally, attempts were also made to study the microstructure of microwave-cured materials by using a scanning electron microscope in order to analyze the morphology of cured specimens.