Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

Delay reduction in courts of justice – possibilities and challenges of process improvement in professional public organizations

Delays in the justice system have been undermining the functioning and performance of the court system all over the world for decades. Despite the widespread concern about delays, the solutions have not kept up with the growth of the problem. The delay problem existing in the justice courts processes is a good example of the growing need and pressure in professional public organizations to start improving their business process performance.This study analyses the possibilities and challenges of process improvement in professional public organizations. The study is based on experiences gained in two longitudinal action research improvement projects conducted in two separate Finnish law instances; in the Helsinki Court of Appeal and in the Insurance Court. The thesis has two objectives. First objective is to study what kinds of factors in court system operations cause delays and unmanageable backlogs and how to reduce and prevent delays. Based on the lessons learned from the case projects the objective is to give new insights on the critical factors of process improvement conducted in professional public organizations. Four main areas and factors behind the delay problem is identified: 1) goal setting and performance measurement practices, 2) the process control system, 3) production and capacity planning procedures, and 4) process roles and responsibilities. The appropriate improvement solutions include tools to enhance project planning and scheduling and monitoring the agreed time-frames for different phases of the handling process and pending inventory. The study introduces the identified critical factors in different phases of process improvement work carried out in professional public organizations, the ways the critical factors can be incorporated to the different stages of the projects, and discusses the role of external facilitator in assisting process improvement work and in enhancing ownership towards the solutions and improvement. The study highlights the need to concentrate on the critical factors aiming to get the employees to challenge their existing ways of conducting work, analyze their own processes, and create procedures for diffusing the process improvement culture instead of merely concentrating of finding tools, techniques, and solutions appropriate for applications from the manufacturing sector

Software for design of experiments and response modelling of cake ltration applications

Filtration is a widely used unit operation in chemical engineering. The huge variation in the properties of materials to be ltered makes the study of ltration a challenging task. One of the objectives of this thesis was to show that conventional ltration theories are di cult to use when the system to be modelled contains all of the stages and features that are present in a complete solid/liquid separation process. Furthermore, most of the ltration theories require experimental work to be performed in order to obtain critical parameters required by the theoretical models. Creating a good overall understanding of how the variables a ect the nal product in ltration is somewhat impossible on a purely theoretical basis. The complexity of solid/liquid separation processes require experimental work and when tests are needed, it is advisable to use experimental design techniques so that the goals can be achieved. The statistical design of experiments provides the necessary tools for recognising the e ects of variables. It also helps to perform experimental work more economically. Design of experiments is a prerequisite for creating empirical models that can describe how the measured response is related to the changes in the values of the variable. A software package was developed that provides a ltration practitioner with experimental designs and calculates the parameters for linear regression models, along with the graphical representation of the responses. The developed software consists of two software modules. These modules are LTDoE and LTRead. The LTDoE module is used to create experimental designs for di erent lter types. The lter types considered in the software are automatic vertical pressure lter, double-sided vertical pressure lter, horizontal membrane lter press, vacuum belt lter and ceramic capillary action disc lter. It is also possible to create experimental designs for those cases where the variables are totally user de ned, say for a customized ltration cycle or di erent piece of equipment. The LTRead-module is used to read the experimental data gathered from the experiments, to analyse the data and to create models for each of the measured responses. Introducing the structure of the software more in detail and showing some of the practical applications is the main part of this thesis. This approach to the study of cake ltration processes, as presented in this thesis, has been shown to have good practical value when making ltration tests.

Indicent reporting in shipping. Experiences and best practices for the Baltic Sea

Incident and near miss reporting is one of the proactive tools of safety management. By analyzing incidents and near misses and by corrective actions, severe accidents can potentially be avoided. Near miss and incident reporting is widely used in many riskprone industries such as aviation or chemical industry. In shipping incident and near miss reporting is required by the mandatory safety management system International Safety Management Code (ISM Code). However, in several studies the conclusion has been that incidents are reported poorly in the shipping industry. The aim of this report is to highlight the best practices for incident reporting in shipping and to support the shipping industry in the better utilization of incident reporting information. The study consists of three parts: 1) voluntary, shared reporting systems in shipping (international experiences), 2) interview study at four shipping companies in Sweden and in Finland (best practices), 3) expert workshop on incident reporting (problems and solutions). Preconditions for a functional reporting system are an existing no blame culture, commitment of the top management, feedback, good communication, training and an easy-to-use system. Although preconditions are met, problems can still appear, for example due to psychological, interpersonal or nationality-related reasons. In order to keep the reporting system functioning, the shipping company must be committed to maintain and develop the system and to tackle the problems. The whole reporting process from compiling, handling and analyzing a report, creating corrective actions and implementing them has to be handled properly in order to gain benefits from the reporting system. In addition to avoiding accidents, the functional reporting system can also offer other benefits by increasing safety awareness, by improving the overall safety and working conditions onboard, by enhancing team work and communication onboard and between ships and the land-based organization of shipping companies. Voluntary shared reporting systems are supported in the shipping industry in principle, but their development in the Baltic Sea is still in its infancy and the potential benefits of sharing the reports have not been realized. On the basis of this study we recommend that a common reporting system be developed for the Baltic Sea area which all the ships operating in the area could use regardless of their flag. Such a wider system could prevent some of the problems related to the current national systems. There would be more incident cases available in the database and this would support anonymity and thus encourage shipping companies to report to a shared database more frequently. A shared reporting system would contribute to the sharing of experiences and to the wider use of incident information in the shipping industry.

Software for design of experiments and response modelling of cake filtration applications

Filtration is a widely used unit operation in chemical engineering. The huge variation in the properties of materials to be ltered makes the study of ltration a challenging task. One of the objectives of this thesis was to show that conventional ltration theories are di cult to use when the system to be modelled contains all of the stages and features that are present in a complete solid/liquid separation process. Furthermore, most of the ltration theories require experimental work to be performed in order to obtain critical parameters required by the theoretical models. Creating a good overall understanding of how the variables a ect the nal product in ltration is somewhat impossible on a purely theoretical basis. The complexity of solid/liquid separation processes require experimental work and when tests are needed, it is advisable to use experimental design techniques so that the goals can be achieved. The statistical design of experiments provides the necessary tools for recognising the e ects of variables. It also helps to perform experimental work more economically. Design of experiments is a prerequisite for creating empirical models that can describe how the measured response is related to the changes in the values of the variable. A software package was developed that provides a ltration practitioner with experimental designs and calculates the parameters for linear regression models, along with the graphical representation of the responses. The developed software consists of two software modules. These modules are LTDoE and LTRead. The LTDoE module is used to create experimental designs for di erent lter types. The lter types considered in the software are automatic vertical pressure lter, double-sided vertical pressure lter, horizontal membrane lter press, vacuum belt lter and ceramic capillary action disc lter. It is also possible to create experimental designs for those cases where the variables are totally user de ned, say for a customized ltration cycle or di erent piece of equipment. The LTRead-module is used to read the experimental data gathered from the experiments, to analyse the data and to create models for each of the measured responses. Introducing the structure of the software more in detail and showing some of the practical applications is the main part of this thesis. This approach to the study of cake ltration processes, as presented in this thesis, has been shown to have good practical value when making ltration tests.

Immediate Decision-Making and Information Needs in Intensive Care Coordination

The aim of this study was to develop a theoretical model for information integration to support the deci¬sion making of intensive care charge nurses, and physicians in charge – that is, ICU shift leaders. The study focused on the ad hoc decision-making and immediate information needs of shift leaders during the management of an intensive care unit’s (ICU) daily activities. The term ‘ad hoc decision-making’ was defined as critical judgements that are needed for a specific purpose at a precise moment with the goal of ensuring instant and adequate patient care and a fluent flow of ICU activities. Data collection and research analysis methods were tested in the identification of ICU shift leaders’ ad hoc decision-making. Decision-making of ICU charge nurses (n = 12) and physicians in charge (n = 8) was observed using a think-aloud technique in two university-affiliated Finnish ICUs for adults. The ad hoc decisions of ICU shift leaders were identified using an application of protocol analysis. In the next phase, a structured online question¬naire was developed to evaluate the immediate information needs of ICU shift leaders. A national survey was conducted in all Finnish, university-affiliated hospital ICUs for adults (n = 17). The questionnaire was sent to all charge nurses (n = 515) and physicians in charge (n = 223). Altogether, 257 charge nurses (50%) and 96 physicians in charge (43%) responded to the survey. The survey was also tested internationally in 16 Greek ICUs. From Greece, 50 charge nurses out of 240 (21%) responded to the survey. A think-aloud technique and protocol analysis were found to be applicable for the identification of the ad hoc decision-making of ICU shift leaders. During one day shift leaders made over 200 ad hoc decisions. Ad hoc decisions were made horizontally, related to the whole intensive care process, and vertically, concerning single intensive care incidents. Most of the ICU shift leaders’ ad hoc decisions were related to human resources and know-how, patient information and vital signs, and special treatments. Commonly, this ad hoc decision-making involved several multiprofessional decisions that constituted a bundle of immediate decisions and various information needs. Some of these immediate information needs were shared between the charge nurses and the physicians in charge. The majority of which concerned patient admission, the organisation and management of work, and staff allocation. In general, the information needs of charge nurses were more varied than those of physicians. It was found that many ad hoc deci-sions made by the physicians in charge produced several information needs for ICU charge nurses. This meant that before the task at hand was completed, various kinds of information was sought by the charge nurses to support the decision-making process. Most of the immediate information needs of charge nurses were related to the organisation and management of work and human resources, whereas the information needs of the physicians in charge mainly concerned direct patient care. Thus, information needs differ between professionals even if the goal of decision-making is the same. The results of the international survey confirmed these study results for charge nurses. Both in Finland and in Greece the information needs of charge nurses focused on the organisation and management of work and human resources. Many of the most crucial information needs of Finnish and Greek ICU charge nurses were common. In conclusion, it was found that ICU shift leaders make hundreds of ad hoc decisions during the course of a day related to the allocation of resources and organisation of patient care. The ad hoc decision-making of ICU shift leaders is a complex multi-professional process, which requires a lot of immediate information. Real-time support for information related to patient admission, the organisation and man¬agement of work, and allocation of staff resources is especially needed. The preliminary information integration model can be applied when real-time enterprise resource planning systems are developed for intensive care daily management

Character transformations and their functional significance as a key to the evolution of hystricognath Rodentia

Hystricognathi represent a monophyletic taxon within Rodentia. Since phylogenetically analyzed morphological systems are essential for revealing evolutionary processes, this study identifies evolutionary character transformations on the stem lineage of Hystricognathi as derived from the author's own work and the literature. Data so far indicate that evolutionary transformations in the rostral head region, the loss of tactile ability in the outer nasal skin and the mobile arrangement of the associated cartilage, were allied with a switch from omnivorous to herbivorous and fiber-rich nutrition. Additional character transformations in the skull assist in digesting such food. Structures associated with reproduction and placentation show a remarkable pro portion of derived character conditions: the chorioallantoic placenta has a ring-shaped organization and growth structure which optimizes the capacity for passive diffusion, a subplacenta occurred as a specialized region responsible for placental invasion and the inverted yolk sac facilitates substance exchange with the main placenta. Finally, precocial newborns evolved as a derived condition within Rodentia. All things considered, a mode of reproduction is indicated, which does not demand excessive additional energy intake by the mother and is in accordance with her low energetic diet. Hystricognathi possess major character transformations that represent prerequisites for their successful radiation at the time when more open ecosystems and grasslands evolved during Earth history. The analysis resulted in the reconstruction of a life-near picture of the hystricognath stem species pattern with high explanatory power in terms of changes in space and time and their interdependence with biodiversity.

Hezbollah and Taliban - hybrid adversaries in contemporary conflicts? : views on an adversary from the United States Armed Forces perspective

The experiences of the United States Armed Forces of the wars in Iraq and Afghanistan and Israel Defense Forces in the Second Lebanon War resulted a new term to surface called “hybrid warfare”. It was to describe the complexity of today’s battlefield. The term “hy-brid warfare” was never officially defined nor is it today. The updated version of the US ARMY Field Manual 3-0: Operations (Change 1) from February 22, 2011, introduced and defined “hybrid threat” and thus opened the discussion for hybrid adversary. In this thesis a model is introduced according to which any organization, group or an ad-versary can be examined and evaluated to see whether it qualifies as a hybrid adversary. It is demonstrated by the example of Hezbollah, which is recognized as the best example of an organization utilizing “hybrid warfare” and subsequently categorizing as a hybrid adver-sary. The model will be tested with Afghan Taliban to see whether both the model works and Taliban qualifies as a hybrid adversary or not. According to the model used in this thesis, it is concluded that Taliban does not meet the standards of a hybrid adversary, but with acquisition of standoff weapons it would quickly qualify as one. The model proved to work, and it could be used as a tool by intelligence of-ficers for estimating the threat levels of any group or identifying those groups that are al-ready or are about to develop into a hybrid adversary.

Empirical study on the adoption, use and effects of cloud-based testing

Cloud computing is a practically relevant paradigm in computing today. Testing is one of the distinct areas where cloud computing can be applied. This study addressed the applicability of cloud computing for testing within organizational and strategic contexts. The study focused on issues related to the adoption, use and effects of cloudbased testing. The study applied empirical research methods. The data was collected through interviews with practitioners from 30 organizations and was analysed using the grounded theory method. The research process consisted of four phases. The first phase studied the definitions and perceptions related to cloud-based testing. The second phase observed cloud-based testing in real-life practice. The third phase analysed quality in the context of cloud application development. The fourth phase studied the applicability of cloud computing in the gaming industry. The results showed that cloud computing is relevant and applicable for testing and application development, as well as other areas, e.g., game development. The research identified the benefits, challenges, requirements and effects of cloud-based testing; and formulated a roadmap and strategy for adopting cloud-based testing. The study also explored quality issues in cloud application development. As a special case, the research included a study on applicability of cloud computing in game development. The results can be used by companies to enhance the processes for managing cloudbased testing, evaluating practical cloud-based testing work and assessing the appropriateness of cloud-based testing for specific testing needs.

Building the ‘Digital’ Library: Routes to Managing our Institutional Digital Learning, Teaching and Research Assets Together

Presentation at Open Repositories 2014, Helsinki, Finland, June 9-13, 2014

DSpace 2014: JSPUI reborn and research management

Presentation at Open Repositories 2014, Helsinki, Finland, June 9-13, 2014

Opportunities, constraints and constrained opportunities – A study on mothers’ working time patterns in 22 European countries

The aim of this study was to analyse mothers’ working time patterns across 22 European countries. The focus was on three questions: how much mothers prefer to work, how much they actually work, and to what degree their preferred and actual working times are (in)consistent with each other. The focus was on cross-national differences in mothers’ working time patterns, comparison of mothers’ working times to that of childless women and fathers, as well as on individual- and country-level factors that explain the variation between them. In the theoretical background, the departure point was an integrative theoretical approach where the assumption is that there are various kinds of explanations for the differences in mothers’ working time patterns – namely structural, cultural and institutional – , and that these factors are laid in two levels: individual- and country-levels. Data were extracted from the European Social Survey (ESS) 2010 / 2011. The results showed that mothers’ working time patterns, both preferred and actual working times, varied across European countries. Four clusters were formed to illustrate the differences. In the full-time pattern, full-time work was the most important form of work, leaving all other working time forms marginal. The full-time pattern was perceived in terms of preferred working times in Bulgaria and Portugal. In polarised pattern countries, fulltime work was also important, but it was accompanied by a large share of mothers not working at all. In the case of preferred working times, many Eastern and Southern European countries followed it whereas in terms of actual working times it included all Eastern and Southern European countries as well as Finland. The combination pattern was characterised by the importance of long part-time hours and full-time work. It was the preferred working time pattern in the Nordic countries, France, Slovenia, and Spain, but Belgium, Denmark, France, Norway, and Sweden followed it in terms of actual working times. The fourth cluster that described mothers’ working times was called the part-time pattern, and it was illustrated by the prevalence of short and long part-time work. In the case of preferred working times, it was followed in Belgium, Germany, Ireland, the Netherlands and Switzerland. Besides Belgium, the part-time pattern was followed in the same countries in terms of actual working times. The consistency between preferred and actual working times was rather strong in a majority of countries. However, six countries fell under different working time patterns when preferred and actual working times were compared. Comparison of working mothers’, childless women’s, and fathers’ working times showed that differences between these groups were surprisingly small. It was only in part-time pattern countries that working mothers worked significantly shorter hours than working childless women and fathers. Results therefore revealed that when mothers’ working times are under study, an important question regarding the population examined is whether it consists of all mothers or only working mothers. Results moreover supported the use of the integrative theoretical approach when studying mothers’ working time patterns. Results indicate that mothers’ working time patterns in all countries are shaped by various opportunities and constraints, which are comprised of structural, cultural, institutional, and individual-level factors.

Sleep patterns and sleep-related complaints of Brazilian interstate bus drivers

Sleep-related complaints have become a highlight for physicians as well as public health administrators. Studies of sleep patterns and sleep-related complaints of shift workers have been useful in minimizing reduction in the quality of life due to the warping of the sleep-wake cycle. The objective of the present study was to assess patterns of sleep, sleep-related complaints as well as physical activity and scoring rates for depression and anxiety in interstate bus drivers. Data were obtained with a sleep questionnaire, with the Beck inventory for depression, and the State-Trait Anxiety Inventory (STAI). A total of 400 interstate bus drivers from the northern, southern, central-western and south-eastern regions of Brazil were interviewed. Sixty percent of the subjects interviewed presented at least one sleep-related complaint, 16% admitted to have dozed at the wheel while on duty, and 41% stated that they exercised on a regular basis. Other sleep disturbance complaints reported were: sleep latency 29'17"; physical fatigue, 59.8%; mental fatigue, 45.4%; sleepiness, 25.8%; irritability, 20.6%; insomnia, 37.5%, respiratory disturbances, 19.25% and snoring, 20.75%. Scores for anxiety and depression were not in the pathological range. The present data reinforce the view that bus drivers are generally discontent with shift work and its effects on sleep. Consequently, it is very important to establish an appropriate work schedule for drivers, besides implementing photo-therapy and physical activities in order to minimize sleepiness when driving.

Operations management in courts of justice - outlining functions, challenges and development possibilities

Professional services are an increasingly important group of economy and related to them there has been identified a fairly new concept called professional service operations management. However professional service operations management is still quite under-researched area which needs further research especially in specific contexts. This study aims to respond to that need by examining courts as an environment for operations management. As a result there is a preliminary structured description of what operations management is and could be in courts. The study also aims to inspire and tentatively classify possible areas for future research. Courts are examined based on three common perspectives for typical characteristics of professional services which can be identified in literature: the nature of customer role, the nature of professional work and the nature of process and product. The examination is based on research data from several research projects conducted in Finland and other European countries. Based on the examination it can be said that the operational environment of courts is highly complex and demanding because the case as an object of operations management tasks is challenging, the process sets strict requirements to the handling of the cases and the workforce is hard to direct.

Psychometric properties of the Portuguese version of the State-Trait Anxiety Inventory applied to college students: factor analysis and relation to the Beck Depression Inventory

The psychometric properties of the Portuguese version of the trait form of the State-Trait Anxiety Inventory (STAI-T) and its relation to the Beck Depression Inventory (BDI) were evaluated in a large Brazilian college student sample containing 845 women and 235 men. STAI-T scores tended to be higher for women, singles, those who work, and subjects under 30 years. Factor analysis of the STAI-T for total sample and by gender yielded two factors: the first representing a mood dimension and the second being related to worrying or cognitive aspects of anxiety. In order to study the relation between anxiety and depression measures, factor analysis of the combination of the 21 BDI items and the 20 STAI-T items was also carried out. The analysis resulted in two factors that were analyzed according to the tripartite model of anxiety and depression. Most of the BDI items (measuring positive affectivity and nonspecific symptoms of depression) were loaded on the first factor and four STAI-T items that measure positive affectivity. The remaining STAI-T items, all of them measuring negative affect, remained in the second factor. Thus, factor 1 represents a depression dimension and factor 2 measures a mood-worrying dimension. The findings of this study suggest that, although widely used as an anxiety scale, the STAI-T in fact measures mainly a general negative affect.