Histogram feature-based Fisher linear discriminant for face detection

The face pattern is described by pairs of template-based histogram and Fisher projection orientation under the framework of AdaBoost learning in this paper. We assume that a set of templates are available first. To avoid making strong assumptions about distributional structure while still retaining good properties for estimation, the classical statistical model, histogram, is used to summarize the response of each template. By introducing a novel “Integral Histogram Image”, we can compute histogram rapidly. Then, we turn to Fisher linear discriminant for each template to project histogram from d-dimensional subspace to one-dimensional subspace. Best features, used to describe face pattern, are selected by AdaBoost learning. The results of experiments demonstrate that the selected features are much more powerful to represent the face pattern than the simple rectangle features used by Viola and Jones and some variants.     

基于向量组的Fisher线性鉴别分析方法

提出了一种基于向量组的Fisher线性鉴别分析方法。该方法先将原始的高维向量分割为低维的子向量组,再对向量组运用Fisher线性鉴别分析。这种处理方法,不但能够解决任意高维下的小样本问题,而且通过选择恰当的子向量维数,可以从向量中抽取出最有效的特征值。此外,基于向量组的Fisher线性鉴别分析是Fisher线性鉴别分析和二维Fisher线性鉴别分析的进一步推广。     

An empirical comparison and evaluation of minority oversampling techniques on a large number of imbalanced datasets

Learning and mining from imbalanced datasets gained increased interest in recent years. One simple but efficient way to increase the performance of standard machine learning techniques on imbalanced datasets is the synthetic generation of minority samples. In this paper, a detailed, empirical comparison of 85 variants of minority oversampling techniques is presented and discussed involving 104 imbalanced datasets for evaluation. The goal of the work is to set a new baseline in the field, determine the oversampling principles leading to the best results under general circumstances, and also give guidance to practitioners on which techniques to use with certain types of datasets.     

Knowledge management in supply chain: An empirical study from France

Supply chain management has assumed a leading operations strategy position in both manufacturing and service industries, and over the past 10 years companies have seriously implemented supply chain management strategies in their organizations. Knowledge management (KM) is a major enabler of supply chain management, and is a critical element in information intensive and multi-cultured enterprise environments. Realizing the importance of knowledge management in supply chain (SC), an attempt has been made in this paper to propose a conceptual framework for KM in SC and to validate the framework with help of an empirical study conducted with French companies. Finally, a summary of findings and conclusions is presented for KM in SC.     

Code smells as system-level indicators of maintainability: An empirical study

Context

Code smells are manifestations of design flaws that can degrade code maintainability. So far, no research has investigated if these indicators are useful for conducting system-level maintainability evaluations.

Aim

The research in this paper investigates the potential of code smells to reflect system-level indicators of maintainability.

Method

We evaluated four medium-sized Java systems using code smells and compared the results against previous evaluations on the same systems based on expert judgment and the Chidamber and Kemerer suite of metrics. The systems were maintained over a period of up to 4 weeks. During maintenance, effort (person-hours) and number of defects were measured to validate the different evaluation approaches.

Results

Most code smells are strongly influenced by size; consequently code smells are not good indicators for comparing the maintainability of systems differing greatly in size. Also, from the comparison of the different evaluation approaches, expert judgment was found as the most accurate and flexible since it considered effects due to the system's size and complexity and could adapt to different maintenance scenarios.

Conclusion

Code smell approaches show promise as indicators of the need for maintenance in a way that other purely metric-based approaches lack.     

Fisher scoring: An interpolation family and its Monte Carlo implementations

The Fisher scoring method is widely used for likelihood maximization, but its application can be difficult in situations where the expected information matrix is not available in closed form or when parameters have constraints. In this paper, we describe an interpolation family that generalizes the Fisher scoring method and propose a general Monte Carlo approach that makes these generalized methods also applicable in such situations. With this approach, random samples are generated from the iteratively estimated models and used to provide estimates of the expected information. As a result, the likelihood function can be optimized by repeatedly solving weighted linear regression problems. Specific extensions of this general approach to fitting multivariate normal mixtures and to fitting mixed-effects models with a single discrete random effect are also described. Numerical studies show that the proposed algorithms are fast and reliable to use, as compared with the classical expectation-maximization algorithm.     

Representing time intervals in a two-dimensional space: An empirical study

ObjectiveInstead of the linear model (LM), time intervals can be represented by a two-dimensional (2D) model, which is called the triangular model (TM). Although the TM has been introduced for decades and applied in some areas, there still a lack of empirical studies on its usability. To fill this gap, this study aims to evaluate how people perform when using the TM to answer questions on time intervals, in comparison with using the traditional LM.MethodAround 250 novice participants took part in the experiment, which consisted of a video training, a pretest and posttest. The video training introduced the basic knowledge of temporal relations and the two representations. The pretest allowed participants to practice the knowledge they have learned and receive feedbacks of the answers. In the posttest, participants' accuracy and speed when answering the questions were recorded for analysis. The results of using the TM and the LM were compared in pairs. The null hypothesis is that the participants produce equal results with the two models.ResultThe results showed that the participants scored better and spent less time when answering questions with the TM, which rejected the null hypothesis. Moreover, the score and speed when they used the TM did decline in the questions containing a larger number of intervals. In contrast, the score and accuracy when they used the LM declined when questions containing a large number of intervals.Conclusion

• •The TM is easy to learn. After a 20-min training, novice participants can use it to solve questions and produce satisfactory result.
• •The TM is easy and efficient for visual queries of time intervals.
• •The TM is easy to use for handling a large number of intervals.

Implication

• •The TM can be widely applied in analysing time intervals and linear data.
• •Tools implementing the TM can be learned and used by novice users.

     

Class movement and re-location: An empirical study of Java inheritance evolution

Inheritance is a fundamental feature of the Object-Oriented (OO) paradigm. It is used to promote extensibility and reuse in OO systems. Understanding how systems evolve, and specifically, trends in the movement and re-location of classes in OO hierarchies can help us understand and predict future maintenance effort. In this paper, we explore how and where new classes were added as well as where existing classes were deleted or moved across inheritance hierarchies from multiple versions of four Java systems. We observed first, that in one of the studied systems the same set of classes was continuously moved across the inheritance hierarchy. Second, in the same system, the most frequent changes were restricted to just one sub-part of the overall system. Third, that a maximum of three levels may be a threshold when using inheritance in a system; beyond this level very little activity was observed, supporting earlier theories that, beyond three levels, complexity becomes overwhelming. We also found evidence of ‘collapsing’ hierarchies to bring classes up to shallower levels. Finally, we found that larger classes and highly coupled classes were more frequently moved than smaller and less coupled classes. Statistical evidence supported the view that larger classes and highly coupled classes were less cohesive than smaller classes and lowly coupled classes and were thus more suitable candidates for being moved (within an hierarchy).     

Essence of kernel Fisher discriminant: KPCA plus LDA

In this paper, the method of kernel Fisher discriminant (KFD) is analyzed and its nature is revealed, i.e., KFD is equivalent to kernel principal component analysis (KPCA) plus Fisher linear discriminant analysis (LDA). Based on this result, a more transparent KFD algorithm is proposed. That is, KPCA is first performed and then LDA is used for a second feature extraction in the KPCA-transformed space. Finally, the effectiveness of the proposed algorithm is verified using the CENPARMI handwritten numeral database.     

Comparing traditional conceptual modeling with ontology-driven conceptual modeling: An empirical study

This paper conducts an empirical study that explores the differences between adopting a traditional conceptual modeling (TCM) technique and an ontology-driven conceptual modeling (ODCM) technique with the objective to understand and identify in which modeling situations an ODCM technique can prove beneficial compared to a TCM technique. More specifically, we asked ourselves if there exist any meaningful differences in the resulting conceptual model and the effort spent to create such model between novice modelers trained in an ontology-driven conceptual modeling technique and novice modelers trained in a traditional conceptual modeling technique. To answer this question, we discuss previous empirical research efforts and distill these efforts into two hypotheses. Next, these hypotheses are tested in a rigorously developed experiment, where a total of 100 students from two different Universities participated. The findings of our empirical study confirm that there do exist meaningful differences between adopting the two techniques. We observed that novice modelers applying the ODCM technique arrived at higher quality models compared to novice modelers applying the TCM technique. More specifically, the results of the empirical study demonstrated that it is advantageous to apply an ODCM technique over an TCM when having to model the more challenging and advanced facets of a certain domain or scenario. Moreover, we also did not find any significant difference in effort between applying these two techniques. Finally, we specified our results in three findings that aim to clarify the obtained results.     

Some multiclass Fisher feature selection algorithms and their comparison with Karhunen-Loève algorithms

The feature selection problem as a task of a transformation of an initial pattern space into a new space, optimal with respect to the discriminatory features is described. Transformation optimizations are realized according to the measures which may be included in the broadly understood group of Fisher measures. In particular, the use of some conception of interclass scatter matrix calculation allows us to obtain different variations of many-class Fisher measures. Finally, a theoretical comparison of some properties of the suggested Fisher transformations with other transformations based on the Karhunen-Loève expansion is presented.     

Median Fisher Discriminator: a robust feature extraction method with applications to biometrics

In existing Linear Discriminant Analysis (LDA) models, the class population mean is always estimated by the class sample average. In small sample size problems, such as face and palm recognition, however, the class sample average does not suffice to provide an accurate estimate of the class population mean based on a few of the given samples, particularly when there are outliers in the training set. To overcome this weakness, the class median vector is used to estimate the class population mean in LDA modeling. The class median vector has two advantages over the class sample average: (1) the class median (image) vector preserves useful details in the sample images, and (2) the class median vector is robust to outliers that exist in the training sample set. In addition, a weighting mechanism is adopted to refine the characterization of the within-class scatter so as to further improve the robustness of the proposed model. The proposed Median Fisher Discriminator (MFD) method was evaluated using the Yale and the AR face image databases and the PolyU (Polytechnic University) palmprint database. The experimental results demonstrated the robustness and effectiveness of the proposed method.     

High-precision control of linear actuators incorporating acceleration sensing

This paper presents the application of the acceleration sensor in the enhancement of the performance of high-precision motion tracking linear actuators which are based on permanent magnet linear motors (PMLM). A feedforward–feedback control structure is developed which harness effectively the acceleration measurements made available. It utilises a linear full-state feedback controller and an iterative learning feedforward controller (ILC). Experimental results show the acceleration feedback can improve the tracking performance and learning convergence of the control system.     

Personality,emotional intelligence and work preferences in software engineering: An empirical study

ContextThere is an increasing awareness among Software Engineering (SE) researchers and practitioners that more focus is needed on understanding the engineers developing software. Previous studies show significant associations between the personalities of software engineers and their work preferences.ObjectiveVarious studies on personality in SE have found large, small or no effects and there is no consensus on the importance of psychometric measurements in SE. There is also a lack of studies employing other psychometric instruments or using larger datasets. We aim to evaluate our results in a larger sample, with software engineers in an earlier state of their career, using advanced statistics.MethodAn operational replication study where extensive psychometric data from 279 master level students have been collected in a SE program at a Swedish University. Personality data based on the Five-Factor Model, Trait Emotional Intelligence Questionnaire and Self-compassion have been collected. Statistical analysis investigated associations between psychometrics and work preferences and the results were compared to our previous findings from 47 SE professionals.ResultsAnalysis confirms existence of two main clusters of software engineers; one with more “intense” personalities than the other. This corroborates our earlier results on SE professionals. The student data also show similar associations between personalities and work preferences. However, for other associations there are differences due to the different population of subjects. We also found connections between the emotional intelligence and work preferences, while no associations were found for self-compassion.ConclusionThe associations can help managers to predict and adapt projects and tasks to available staff. The results also show that the Emotional Intelligence instrument can be predictive. The research methods and analytical tools we employ can detect subtle associations and reflect differences between different groups and populations and thus can be important tools for future research as well as industrial practice.     

Client-side selection of replicated web services: An empirical assessment

Replicating web services over physically distributed servers can offer client applications a number of QoS benefits, including higher availability and reduced response time. However, selecting the “best” service replica to invoke at the client-side is not a trivial task, as this requires taking into account factors such as local and external network conditions, and the servers’ current workload. This paper presents an empirical assessment of five representative client-side service selection policies for accessing replicated web services. The assessment measured the response time obtained with each of the five policies, at two different client configurations, when accessing a world-wide replicated service with four replicas located in three continents. The assessment’s results were analyzed both quantitatively and qualitatively. In essence, the results show that, in addition to the QoS levels provided by the external network and the remote servers, characteristics of the local client environment can have a significant impact on the performance of some of the policies investigated. In this regard, the paper presents a set of guidelines to help application developers in identifying a server selection policy that best suits a particular service replication scenario.     

Skills required for participating in CMC courses: An empirical study

The development of new communication technologies and their applications has opened a broad spectrum of options to promote learning, of which a significant one is CMC – Computer-Mediated Communication. Yet, students use this medium to a relatively small extent. Our premise is that the use of these technologies depends on the level of skills and expertise of the individuals using them. The purpose of this study is to determine whether taking a Computer Literacy and Applications course (CLA) before taking CMC courses has an impact on students' participation in CMC courses and on the effective use of various CMC tools. The sample consisted of 153 students enrolled in CMC courses in the Social Sciences in the Spring 2001/2 and Fall 2002/3 semesters. Participants were divided into two groups: 80 students who had taken CLA, and 73 students who had not. A questionnaire to assess the impact of CLA on participation in CMC courses was administered to the two groups. The results of the study indicate that participation in CMC courses is more intensive in the group that had taken CLA. Based on these findings, we recommend mandating a CLA course, like the one developed at the Open University of Israel (OUI), before taking CMC courses.     

Locally linear discriminant embedding: An efficient method for face recognition

In this paper an efficient feature extraction method named as locally linear discriminant embedding (LLDE) is proposed for face recognition. It is well known that a point can be linearly reconstructed by its neighbors and the reconstruction weights are under the sum-to-one constraint in the classical locally linear embedding (LLE). So the constrained weights obey an important symmetry: for any particular data point, they are invariant to rotations, rescalings and translations. The latter two are introduced to the proposed method to strengthen the classification ability of the original LLE. The data with different class labels are translated by the corresponding vectors and those belonging to the same class are translated by the same vector. In order to cluster the data with the same label closer, they are also rescaled to some extent. So after translation and rescaling, the discriminability of the data will be improved significantly. The proposed method is compared with some related feature extraction methods such as maximum margin criterion (MMC), as well as other supervised manifold learning-based approaches, for example ensemble unified LLE and linear discriminant analysis (En-ULLELDA), locally linear discriminant analysis (LLDA). Experimental results on Yale and CMU PIE face databases convince us that the proposed method provides a better representation of the class information and obtains much higher recognition accuracies.