Feature selection via sensitivity analysis of SVM probabilistic outputs

Feature selection is an important aspect of solving data-mining and machine-learning problems. This paper proposes a feature-selection method for the Support Vector Machine (SVM) learning. Like most feature-selection methods, the proposed method ranks all features in decreasing order of importance so that more relevant features can be identified. It uses a novel criterion based on the probabilistic outputs of SVM. This criterion, termed Feature-based Sensitivity of Posterior Probabilities (FSPP), evaluates the importance of a specific feature by computing the aggregate value, over the feature space, of the absolute difference of the probabilistic outputs of SVM with and without the feature. The exact form of this criterion is not easily computable and approximation is needed. Four approximations, FSPP1-FSPP4, are proposed for this purpose. The first two approximations evaluate the criterion by randomly permuting the values of the feature among samples of the training data. They differ in their choices of the mapping function from standard SVM output to its probabilistic output: FSPP1 uses a simple threshold function while FSPP2 uses a sigmoid function. The second two directly approximate the criterion but differ in the smoothness assumptions of criterion with respect to the features. The performance of these approximations, used in an overall feature-selection scheme, is then evaluated on various artificial problems and real-world problems, including datasets from the recent Neural Information Processing Systems (NIPS) feature selection competition. FSPP1-3 show good performance consistently with FSPP2 being the best overall by a slight margin. The performance of FSPP2 is competitive with some of the best performing feature-selection methods in the literature on the datasets that we have tested. Its associated computations are modest and hence it is suitable as a feature-selection method for SVM applications. Editor: Risto Miikkulainen.     

A case study comparing defect profiles of a reused framework and of applications reusing it

The benefits of software reuse have been studied for many years. Several previous studies have observed that reused software has a lower defect density than newly built software. However, few studies have investigated empirically the reasons for this phenomenon. To date, we have only the common sense observation that as software is reused over time, the fixed defects will accumulate and will result in high-quality software. This paper reports on an industrial case study in a large Norwegian Oil and Gas company, involving a reused Java class framework and two applications that use that framework. We analyzed all trouble reports from the use of the framework and the applications according to the Orthogonal Defect Classification (ODC), followed by a qualitative Root Cause Analysis (RCA). The results reveal that the framework has a much lower defect density in total than one application and a slightly higher defect density than the other. In addition, the defect densities of the most severe defects of the reused framework are similar to those of the applications that are reusing it. The results of the ODC and RCA analyses reveal that systematic reuse (i.e. clearly defined and stable requirements, better design, hesitance to change, and solid testing) lead to lower defect densities of the functional-type defects in the reused framework than in applications that are reusing it. However, the different “nature” of the framework and the applications (e.g. interaction with other software, number and complexity of business logic, and functionality of the software) may confound the causal relationship between systematic reuse and the lower defect density of the reused software. Using the results of the study as a basis, we present an improved overall cause–effect model between systematic reuse and lower defect density that will facilitate further studies and implementations of software reuse.

Slips and falls in a cold climate: underfoot surface, footwear design and worker preferences for preventive measures

Slips and falls and associated outdoor injuries are prevalent in cold climates. The objectives of this field investigation were to describe the consequences of slips and falls on ice and snow and the associated injuries, to assess the risks of various icy and snowy surfaces, to identify design needs of footwear, and to ascertain preventive measure preferences of outdoor workers. The organizations investigated were a newspaper delivery service, a military regiment, mining and construction industries. The results showed that fall events occur most frequently on ice covered with snow. This is due to the difficulty of perceiving hidden risks in order to adjust gait strategies. The professional footwear provided does not provide enough protection against slips and falls. Slip resistant properties are ranked as one of the top requirements by the users. Their most preferred preventive measures are footwear with anti-slip properties and the application of anti-slip materials, such as sand or salt.     

Interventional ultrasound in Europe

"Interventional ultrasound," defined as any diagnostic and therapeutic procedure performed under ultrasound guidance was first introduced in Europe, where its early development took place in Vienna, in Copenhagen, in Italy and in Switzerland. However, many of the applications of interventional ultrasound have been based on important pioneer work using other less-suitable needle guiding methods from the pre-ultrasound era. The European contributions to "interventional ultrasound" have especially been in the development of new puncture equipment, in the dissemination of various biopsy techniques, and draining procedures and, more recently, in the development of many different tissue-ablation techniques. The above contributions, which are outlined in this historical review, have, together with significant contributions from the rest of the world-not least from the United States and Japan-created a most efficient diagnostic as well as therapeutic tool for the benefit of our patients.     

Simultaneous machining of parallel grooves in SnO(2) thin films using a Nd:YAG laser and a kinoform

A kinoform is used to split the beam of a Nd:YAG laser into six beams. The laser beams are, by means of optical fibers and collimating-focusing optics, transferred to the surface to be machined. Thus multiple grooves can be machined simultaneously. For demonstration, 100-μm-wide grooves are simultaneously machined into a SnO(2) thin film deposited on a glass substrate. The resulting grooves are well isolated. This result shows that the technique could be used, for example, in solar cell manufacturing to increase the efficiency of laser scribing.     

The Effect of Subfilter-Scale Physics on Regularization Models

The subfilter-scale (SFS) physics of regularization models are investigated to understand the regularizations’ performance as SFS models. Suppression of spectrally local SFS interactions and conservation of small-scale circulation in the Lagrangian-averaged Navier-Stokes α-model (LANS-α) is found to lead to the formation of rigid bodies. These contaminate the superfilter-scale energy spectrum with a scaling that approaches k ⁺¹ as the SFS spectra is resolved. The Clark-α and Leray-α models, truncations of LANS-α, do not conserve small-scale circulation and do not develop rigid bodies. LANS-α, however, is closest to Navier-Stokes in intermittency properties. All three models are found to be stable at high Reynolds number. Differences between L ² and H ¹ norm models are clarified. For magnetohydrodynamics (MHD), the presence of the Lorentz force as a source (or sink) for circulation and as a facilitator of both spectrally nonlocal large to small scale interactions as well as local SFS interactions prevents the formation of rigid bodies in Lagrangian-averaged MHD (LAMHD-α). LAMHD-α performs well as a predictor of superfilter-scale energy spectra and of intermittent current sheets at high Reynolds numbers. It may prove generally applicable as a MHD-LES.     

Integrated micro Pirani gauge based hermetical package monitoring for uncooled VO x bolometer FPAs

This paper presents the design and fabrication of a micro Pirani gauge using VO _x as the sensitive material for monitoring the pressure inside a hermetical package for micro bolometer focal plane arrays (FPAs). The designed Pirani gauge working in heat dissipating mode was intentionally fabricated using standard MEMS processing which is highly compatible with the FPAs fabrication. The functional layer of the micro Pirani gauge is a VO _x thin film designed as a 100 × 200 μm pixel, suspended 2 μm above the substrate. By modeling of rarefied gas heat conduction using the Extended Fourier’s law, finite element analysis is used to investigate the sensitivity of the pressure gauge. Also the thermal interactions between the micro Pirani gauge and bolometer FPAs are verified. From the fabricated prototype, the measured device TCR is about −0.8% K⁻¹ and the sensitivity about 1.84 × 10⁻³ W K⁻¹ mbar⁻¹.     

Digital Game Dynamics Preferences and Player Types

In this study, we examine digital game preferences by identifying game dynamics, i.e. player–game interaction modes, of 700 contemporary digital games, and players' (N = 1717) desire to play games with specific types of dynamics. Based on statistical analysis of the data, 5 game dynamics preference categories (“assault,” “manage,” “journey,” “care,” and “coordinate”) and 7 player types were revealed. The results show that identifying player types requires including both preferred and undesired game dynamics categories in the analysis. The findings unveil digital gaming as a more multifaceted phenomenon than common stereotypes suggest. The original game preferences model we present in this study can be conceptualized as a complementary approach for motivations to play and player behavior studies.     

Classification of metabolic and respiratory demands in fire fighting activity with extreme workloads

Fire fighting work comprises work tasks requiring an energy yield at maximal or close to maximal levels of the individual. Due to the very nature of fire fighting more complex physiological variables are difficult to measure. We measured metabolic and respiratory responses in 15 male, professional fire fighters during simulated work tasks on a test ground. Work time was on the average 22 min with individual components of work tasks lasting 2-4 min. The mean oxygen consumption for the whole exercise (22 min) was 2.75+/-0.29 l/min. The most demanding work task demanded an oxygen uptake of 3.55+/-0.27 l/min. Corresponding values for respiratory minute volumes were 82+/-14 and 102+/-14l/min, respectively. Heart rates averaged 168+/-12 for the whole test and 179+/-13 beats/min for the heaviest work task. Two new classes for classification of intensive and exhausting, short term physical work are proposed for inclusion in ISO8996 and values for relevant parameters are proposed.     

Predicting dementia from the Mini-Mental State Examination in an elderly population: the role of education

Steroid hormone regulation and cell-type specific expression of the jun-D protooncogene in rat uterus was examined. Adult, ovariectomized rats were injected with progesterone, testosterone, 17beta-estradiol (E2-17beta), 16alpha-estradiol (E2-16alpha), dexamethasone or cycloheximide. Uteri were collected between 0 and 6 h post-treatment. Northern blot analysis of uterine RNA revealed that induction of jun-D was specific for estrogenic steroids, as progesterone and testosterone had no effect on expression of this member of the jun gene family. Treatment with E2-17beta increased jun-D mRNA levels by approx. 5-fold, with expression reaching peak levels at 3 h after treatment and declining thereafter. Administration of E2-16alpha, a short-acting estrogen that does not cause uterine cell proliferation, increased expression of jun-D but with different kinetics compared to the long-acting E2-17beta. The mRNA levels of jun-D increased by 3-fold 1 h after administration of E2-16alpha but declined soon after. Slight induction of jun-D mRNA by dexamethasone was apparent, but to a much lesser extent compared to estrogen. The protein synthesis inhibitor, cycloheximide, did not block jun-D induction, indicating that this is an "immediate early" response. Expression of Jun-D protein was examined by immunohistochemical methods. E2-17beta treatment activated jun-D primarily in the nuclei of luminal and glandular epithelial cells of the endometrium. These results demonstrate that hormonal induction of jun-D is specific for estrogens and that uterine expression of this protooncogene occurs in a cell-type specific manner.