Past PhD theses

This thesis addresses the social challenges of large-scale agile transformations in organisations. While agile methods have proven effective in small teams, extending these practices across entire organisations remains a significant challenge. Expected benefits, such as increased speed and collaboration, are often not realised due to “people challenges” such as resistance to change, scepticism, and insufficient training – issues deeply embedded in organisational culture.
To tackle these challenges, the thesis is structured into three chapters comprising six essays. Chapter I conceptualises large-scale agile transformations from a socio-perspective, framing them as wicked problems. Chapter II develops and designs the Agile Transformation Canvas, a visual inquiry tool designed to support the collaborative and non-linear exploration of wicked problems, specifically addressing the social challenges of scaling agile within organisations. Chapter III extends Chapter II by digitalising the tool, incorporating systems thinking to further account for the wickedness of scaling agile.
By focusing on the cultural and social dimensions of large-scale agile transformations, this research offers both practical tools and theoretical insights to guide organisations in overcoming these challenges, making valuable contributions to both research and practice.

In recent decades, there has been growing interest in location-based services using Global Satellite Navigation Systems (GNSS) such as the Global Positioning System (GPS) or Galileo. These systems require a direct line of sight to the satellites to operate. Although we spend most of our time indoors, there is no standard solution for tracking people or assets in indoor environments. This is mainly because existing Indoor Tracking Systems (ITS) are complex to deploy or offer low accuracy under real-life conditions.
In this thesis, we present five papers divided into three chapters. The first chapter reviews the literature and proposes a unified conceptual model to classify and compare the existing ITS.
In the second chapter, we propose solutions to reduce infrastructure costs while improving the accuracy of ITS. To achieve this, we rely on an inertial-based ITS, accumulating errors over time and leading to drifting trajectories. We propose novel error correction algorithms complemented by fixed Bluetooth Low Energy (BLE) beacons positioned at strategic locations. Our algorithms use a collaborative approach, enabling mobile devices to correct their location estimates by opportunistically exchanging data in a peer-to-peer manner.
The third chapter addresses the challenges of devising, evaluating, and fine-tuning tracking algorithms. Our aim is to speed up research and bring transparency to experiments to guarantee their reproducibility. To achieve this, we propose MobiXIM, a novel framework that aims to standardise the processes for building tracking algorithms. We aim to further research and innovation in ITS by providing open access to all our experimental data.

Health-related data (HRD) is generated in the clinical setting through the use of healthcare information systems by practitioners, and in the personal setting, through the use of digital services. Although HRD has many benefits in the advancement of healthcare, they are considered highly sensitive. This thesis explores and addresses multiple aspects of HRD privacy.

First, we design, implement, and evaluate an efficient and privacy-preserving protocol for collaborative training of decision tree models on distributed biomedical datasets. Our protocol achieves a good balance between model accuracy, efficiency, and privacy, making it suitable for use in the healthcare setting.

Second, we examine shadow HRD: HRD generated and/or processed by individuals by using general-purpose digital tools outside of a professional healthcare information system. These can be overlooked when enforcing data protection laws pertaining to HRD, as they are difficult to detect. We identify and categorize a broad variety of user behaviors that lead to the creation of shadow HRD. We also analyze the privacy policies of popular service providers, and show that although some of them do treat shadow HRD as sensitive data, many do not even acknowledge their collection through their service.

Third, we assess the prevalence of shadow HRD-creating behaviors, users' awareness and concerns with respect to shadow HRD, and the relevant privacy-utility trade-offs. We find that users deem that the benefits of using general-purpose tools for managing their health outweigh the privacy risks to their health data. Furthermore, users rarely resort to privacy protection measures when doing so. This highlights the need for better protections at the service providers' and legal levels.

Industry reports show that many firms still struggle with creating value from their data because data practices remain confined to data experts. Research has yet to explain how companies develop data practices among a widening audience — a capability denoted as democratization — especially how employees can effectively integrate data within their domain expertise. Simultaneously, the role of data governance — providing the framework for managing data as an asset — must shift from a control function focused on data protection to becoming an enabler of innovation. This thesis elucidates how data democratization and data governance co-evolve to scale data practices, through two interrelated streams of research. Through three essays, the first research stream grounds data democratization in Information Systems (IS) research, identifying it as a capability rooted in practice. We emphasize the necessity of integrating both generic and situated data practices to achieve data democratization. We illustrate how data practices are cultivated through situated learning and practice exchange. Through two essays, the second research stream explains how to govern data to achieve both control and innovation. We introduce systems thinking to position data governance at the intersection of data strategy and data operations within a Viable System Model. We describe the reconfiguration of data governance into archetypes that reflect the evolving strategic role of data. Altogether, our findings advance data management research by providing a clearer understanding of how to scale data practices through the interplay between data democratization and data governance, synergistically driving value creation from data.

In the age of digital transformation, enterprises are increasingly recognizing the value of external data, which originates beyond their four walls. Despite the growing number of datasets and their potential value, external data is sourced in an ad-hoc manner without clear guidelines. This leads to inconsistent sourcing decisions, characterized by a lack of clarity on the object of sourcing and the underlying data sourcing practices. Furthermore, the field of data sourcing lacks extensive research, necessitating urgent action within the Information Systems (IS) research community to bridge this gap. Considering the abovementioned research opportunities, this thesis – through three interrelated research streams – provides foundations for, analyzes, and improves data sourcing practices in the enterprise context. The contributions of the first research stream are an external data sourcing taxonomy (Essay 1), which informs sourcing decisions in an enterprise context, and a reference process to source and manage external data, which is accompanied by explicit prescriptions in the form of design principles (Essay 2). The second research stream proposes a use case-driven assessment of open corporate registers (Essay 3) and, building on the subsequent findings, a method to screen, assess, and prepare open data for use in support of companies’ open data activities (Essay 4). Finally, the third research stream reveals and elaborates on three data sourcing practices developed by companies in response to institutional pressures in the sustainability context (Essay 5). Thus, the outcomes of this thesis enable the transition from ad-hoc acquisition to well-informed, professional data sourcing approaches in the enterprise context.

Wearable devices, such as wearable activity trackers (WATs), are increasing in popularity. Although they can help improve a person's quality of life, they also raise serious privacy issues. Although security aspects of WATs have been widely studied (e.g., Bluetooth security, inference of password or biometrics), as well as privacy-related aspects such as users' attitudes and concerns, we lack knowledge about the privacy of WAT users. Indeed, the security aspects that were studied in prior work are not enough to build a realistic adversary model, as these studies focus mostly on communication protocols and not on large-scale data collection. Furthermore, previous work related to data inference by using WATs focuses on only functionalities rather than on privacy (e.g., better monitoring of activity or health to improve user experience). Moreover, these studies focus only on the inference of behavioral patterns (e.g., activities, consumption) or conditions (e.g., diseases), but none of them investigate the inference of users’ personal attributes (e.g., personality, religion, political views). In this thesis, composed of three research papers and a literature review, we contribute to the WAT security \& privacy research field by analyzing how the data of WAT users can be accessed at a large scale by many potential adversaries, by evaluating how such data can be used to infer users' personal attributes and, finally, by proposing privacy enhancing technologies (PETs) to protect their privacy. Concretely, after analyzing the current literature about WAT security \& privacy, we conduct a user-survey study to better understand the WAT user’s behaviors towards data sharing, especially with respect to third-party applications (TPAs) that can easily be used by adversaries to collect data. We then use a rigorous machine-learning approach to evaluate to what extent users’ psychological profiles (Big 5) can be inferred from WAT data, and we discuss the related consequences on the users’ privacy and society as a whole. Finally, to propose effective and likely-to-be-adopted protection mechanisms, we conduct a user-centered design study by using a participatory design methodology before analyzing and evaluating the proposed designs in order.

The growing use of complex Machine Learning (ML) models, especially in critical domains such as healthcare and finance, raises concerns about security, ethics, and comprehensibility. The field of Interpretable Machine Learning (IML) emerged to provide insights into the workings of these models. In this thesis, we cover interpretability methods for both classical ML and deep neural networks (DNNs). In terms of classical ML methods, the significant contribution of our work is MoDE, a novel technique that generates interpretable, low-dimensional visualizations for large-scale, high-dimensional datasets. MoDE stands out by preserving not only inter-data point distances but also correlations and ordinal scores, offering enhanced data understanding. We also evaluate various explanation methods for DNNs, revealing that the selection of the optimal method depends on the specific task. Moreover, we analyze the robustness of these methods against adversarial attacks, highlighting vulnerabilities and proposing novel attack strategies with sparse perturbations. These findings carry implications for enhancing the robustness of explanation methods.

Furthermore, as an application of IML, we apply IML to recommender systems, utilizing graph neural networks to achieve superior recommendation accuracy and interpretability. By leveraging user interaction subgraphs, we are able to provide in-depth interpretations for recommendations. This dual emphasis on accuracy and interpretability holds promise for enhancing recommender systems in practical business scenarios.

Mobile apps are used to support behavior change goals (e.g., stopping a lousy habit, increasing the physical activity frequency, or learning a new skill). Because these apps are pervasive, they are great tools to reach and help people in their self-improvement path towards better habits and well-being. However, users of these apps fail to reach their objectives because they lack the motivation to attain their goals. App creators might help to support motivation by integrating human motivation theories into their designs. Still, not many apps are theory-grounded. To address this issue, we organized our research around a comprehensive human motivation theory: The Self-Determination Theory (SDT). We explored three research streams: (1) developing an artifact that maps market app features to support the SDT Basic Psychological Needs (BPNs); (2) creating an SDT inspired physical activity app that provided empirical evidence that its design supported the BPNs; (3) providing empirical evidence of an SDT inspired appdesign that contributes to increasing the physical activity and motivation of individuals. The results of our studies show that the SDT can inform the design of behavior change app features. Moreover, these SDT inspired features can be used to create a physical activity app that improves individuals’ intrinsic motivation and physical activity level. This thesis results have practical implications for app designers, policymakers, and health practitioners whose interest lies in creating theory-informed and effective behavior change apps.

In view of the ever-changing organisational environment, it is essential to better understand the role that judgment plays in decision-making so that the organisation can function efficiently. This dissertation's overall research focus is on the role of human judgment in strategic, tactical, and operational decision-making inside organisations, as well as in diverse functional domains, specifically demand forecasting and strategic partner selection. Sustainable partner selection for collaborative networked organisations (CNO) is a strategic decision for long-term collaboration with a lasting impact on partners. The focus of this research includes the impact of human judgment for sustainable partner selection for CNO using multi-criteria decision-making methods. The role of human judgment in the demand planning and forecasting function is the focus at the tactical and operational decision-making levels in this research. This research demonstrates the use of human judgment at the operational decision-making level by modelling demand and in the selection of variables, at the tactical decision-making level in model selection in the case of several conflicting error measures, and in the selection of sustainable partners for collaboration at the strategic decision-making level of an organisation. It is found that the structured approaches proposed in the research articles on using human judgment in demand forecasting and partner selection support the decision-making processes at each respective level.

Empirical research on retail investors has attempted to explain online investor subpar performance by inferior decision-making and behavioural biases. The findings on human reasoning have mostly been based on quantitative transaction and account data provided by a single financial institution. The purpose of this study was to explore opportunities to improve investing processes co-created by investors and online investing platform service providers. Qualitative interviews and analysis of live platforms were used to conduct the research.

The analyses and evaluations show that investors have not fully adopted the online platform-provided decision support tools because of issues related to usability, user experience, incompatibility with current investing style and lack of transparency – leading to intuition-based decision-making. Investors requested less complexity and a reduction of information overflow by more meaningful data presentation and interpretation. A considerable proportion of investors had not implemented tenets of portfolio theory and asset allocation, preferring position-focused investing. Investors claimed having learned to mitigate the effect of behavioural biases through experience. A modified design science approach was applied to evaluate the potential for improved decision support adoption. The evaluations show that by alternative design, investors can be persuaded to implement portfolio-level management. Investors are a heterogeneous group of users and modularisation of investing services is suggested in order to match the diversity in investor needs and priorities.

This thesis consists of three scientific publications that use machine learning methods to understand industrial processes that were designed to improve environmental quality. The work contributes to efforts to make numerical models for simulation and process control that can be applied to improve equipment designs and process performance.
Chapter 1 reports the analysis of data obtained from continuous torrefaction trials conducted in an integrated pilot plant. The purpose of the work was to gain insights into the process behavior that might be useful in the conception of a process controller that aims to obtain a product having a high and constant calorific value from heterogeneous raw biomass having a low and fluctuating calorific value.
Chapter 2 reports the development of artificial neural network predictive models of data acquired during 12 months of operation of an anaerobic digester treating food wastes. The purpose of the work was to make a predictive model to map process performance, evaluated in terms of CH4 production, to measurable operational parameters. The models were then assessed for use as decision tools to control the digester feeding rate.
Chapter 3 reports the development of a problem-solving approach and a simulation tool for use in designing upflow anaerobic filters (UAF) for wastewater treatment. A multi-layer perceptron (MLP) artificial neural network model was used to run simulations of different design values of the UAF packing diameter, packing material type and the bioreactor height to diameter ratio. The ranked results of simulation were used to select the optimal design values.

Big old companies surprisingly often fail to seize the opportunities that come with new digital technologies. These well-established “incumbent” organisations do not generally lack financial resources for digital innovation. Nor do they lack ideas for new digital products, processes, and services. Rather, their challenge lies in successfully managing the development of digital ideas, especially the digital ideas that come from employees who have only limited IT skills – so-called “non-IT employees”.

This thesis describes our findings from studying incumbent organisations that support non-IT employees in creating digital innovation. We identify three types of calls for ideas—formal, informal, and mixed—that enable non-IT employees to contribute their knowledge, skills, and creativity to the initiation of digital innovation. We find that the success of these calls for ideas relies on the careful orchestration of organisational competences, managerial processes, and individual practices. We describe how a big old company can leverage these elements to better navigate the initiation of digital innovation.

Incumbent organisations face a challenge to simultaneously explore external technologies and knowledge to develop new revenue sources and exploit their existing assets and capabilities to secure existing revenue streams. Exploration and developing digital innovations call for new skills and flexible processes that allow for experimentation, iteration, and even failure, which can be stand in stark contrast to the accumulated skills and processes that have been refined for exploitation over the years. As a remedy, some organisations are choosing to move digital innovation activities into separate organisational units and toward the peripheries of the organisation. This can however hamper access to the existing knowledge base and capabilities in the rest of the organisation. To mitigate this, previous research has highlighted integration: a collaboration process to combine newly explored knowledge with existing knowledge and assets to develop digital innovation outputs. As yet, few research projects have focused specifically on understanding how distinct integration mechanisms can be used to join exploration and exploitation efforts to maximise the value of digital innovation outputs. Anchored in ambidexterity and digital innovation literature and based on qualitative data, we argue that incumbent organisations need to put in place formal integration mechanisms rather than relying on spontaneous knowledge exchange between explorative and exploitative activities. In terms of organisational units dedicated to digital innovation, we put forth the notion that solely setting up such a unit is unlikely to be sufficient as a performance boosting action but that the integration of such units must be carefully considered and managed at multiple levels, from senior management to individual employees’ behaviours.

In the last few decades, the rise of theoretical computer science led to a growing interest in definability problems on quasi-Polish spaces, a class of topological spaces that was recently introduced by de Brecht. The Scott domain, which is the set of all subsets of the integers equipped with the Scott topology, is universal among these spaces, which makes it our main object of study.

In the first part of the thesis, we show that the partial order induced by reductions via continuous functions on the Borel subsets of the Scott domain is ill-founded and contains infinite antichains. Moreover, we show that these properties already occur at the lowest possible level of topological complexity.

To remedy this situation, we then study the partial order induced by total and relatively continuous relations. We fully characterize this partial order on the Borel subsets of the Scott domain by showing that it is isomorphic to the restriction of the Wadge order on the non-self-dual Borel subsets of the Baire space.

In the second part of the thesis, we are interested in the Decomposability Conjecture, an important problem whose object is the decomposability of Borel functions. Using the question tree machinery developed by Duparc, we present new techniques that allow us to tackle this conjecture from a novel perspective.

In the framework of a joint supervision of doctoral studies with the University of Paris (France).

In recent years, active cyber defence has become an integral part of a State’s toolbox to defend against malicious cyber activities. Complementary to their defensive postures in cyberspace, States have opted for a more active stance in order to identify, prevent and respond to cyber operations. This paradigm shift has also been observed in Switzerland since 2012. This dissertation explores the concept of active cyber defence by comparing and defining the terms, its characteristics and typologies. The scope of this dissertation focuses on the Swiss approach and includes the international and domestic legal framework pertaining to active cyber defence. Furthermore, it depicts the Swiss political governance by scrutinizing the roles and responsibilities of stakeholders at the federal level, the inter-agency collaboration and the decision-making processes. In 2019, a case study was conducted to observe how this political framework was applied during the integrated security exercise which simulated a nationwide cyber crisis. Finally, this dissertation provides a set of recommendations that could improve the Swiss framework surrounding the use of active defensive measures in cyberspace.

In the data-driven enterprise, large quantities of data are automatically processed to detect patterns, generate insights, and fuel business processes and business models alike. However, enterprises striving to become data-driven face multiple challenges. First, while a broadening audience of enterprise stakeholders need to work with data, they largely remain lodged in organizational and technical silos. Second, organizations must comply with strengthened data protection regulations to exploit personal data, which is key to successful data-driven business strategies. This dissertation explores the seemingly contradictory objectives of data protection compliance and increased data usage in enterprises. On the one hand, we investigate. enterprise data catalogs (EDC) as emerging platforms that support data democratization in the enterprise context. Based on insights from over 10 real-world EDC pilot projects in multi-national organizations, we identify emerging data-related roles and analyze how EDCs support their typical data needs and usages. On the other hand, we examine the interplay between data protection regulations and data processing practices, by linking compliance requirements to data management capabilities, embedded into a revisited data life cycle for personal data. In collating these two perspectives, we find that creating transparency on enterprise-wide data assets is a cornerstone of both data protection compliance and data democratization. Our contributions inform the redesign of data management practices and call attention to the need for diligent data documentation as essential building block of the conformable and democratized data-driven enterprise.

Big Data and Analytics (BDA), which describes the ever-growing volumes of digital data and the application of advanced analytics technologies, promises significant value creation opportunities across industries. Even though companies increase their investments, their BDA initiatives fall short of expectations and they struggle to guarantee a return on investments. This lack raises fundamental questions about how companies actually change and adapt their data-related capabilities, especially regarding the quality and use of data, to create value from BDA. Considering these gaps, this thesis – through five interrelated essays – investigates how companies adapt their enterprise data management (EDM) capabilities to create additional business value from BDA. The thesis outcomes show that BDA has profound implications on EDM practices. Companies can create value from their data assets only by managing data quality and data use (e.g., analytics) in correspondence. Establishing the corresponding management structurs and awareness for data require time and commitment over a longer period. Usually, there are many different data initiatives in companies that first need to be consolidated. The thesis outcomes can support practitioners in consolidating these initiatives and formulating a comprehensive EDM approach. First, by gaining a unified view of the analytics initiatives and managing the corresponding capabilities as an integrated enterprise analytics platform (see Essay I). Second, by adapting or establishing data governance designs with a focus on data availability and data monetization (see Essay II and Essay III). Third, by showing practitioners options to run data quality more efficiently with the support of artificial intelligence, specifically machine learning (see Essay IV). Fourth, by supporting the practitioners in setting up an enterprise data catalog to facilitate data use and data democratization for a broad audience within organizations (see Essay V). Thus, the thesis outcomes can support companies in setting up corresponding programs that provide them with a long-term directive to become more data driven.

The operational continuity of critical infrastructures (CIs) is vital for the functioning of modern societies. Yet, these CIs are monitored/managed by an interdependent ecosystem of information systems, exposing CIs to the systemic risk of cascading failures. Consequently, CIs require an information-systems defense capability – i.e., the ability to prevent, detect and respond to information systems’ failure. In order to ensure such a capability, the field of computer & information security develops a myriad of technologies. However, security incidents are caused by inappropriate organizational design and/or human-behavior aspects, at least as often as by inefficient IT design. Following this logic, information systems are apprehended as socio-technical systems constituted by a nexus of technologies (material resources) and human agents (human, and knowledge resources) who employ such technologies. Building on prior research on organizational capabilities and security economics, I explore the organizational design and human behavior aspects that are necessary for CIs to build an information-systems defense capability. Investigating the case of three specific critical infrastructures and their context, I deconstruct this capability into material, human, and knowledge resources, and I explore how they should be acquired to build such a capability. My contributions are threefold: 1) a model that helps to preempt the effect of disruptive technologies on the optimal level of investment in cyber-security, providing a framework in order to select and invest in the most effective technologies; 2) a recruitment framework in order to attract scarce IT-specialists; 3) a model in order to foster tacit-knowledge absorption related to cyber-security. Policy recommendations and a research agenda for future work are presented.

We live in an age where we witness how machines learned to talk, beat us at video games, dream, paint and advance making scientific discoveries. However, having little or no underlying principle that explains the working of these learning systems, leads to a lack of confidence in their efficiency. An inevitable consequence of the lack of trust and certainty, is limited productionalization of machine learning. This notably holds for sensitive domain applications, such as precision medicine, engineering or self-driving vehicles to name a few.

This thesis is an attempt to make a contribution towards more understandable and trustworthy machine learning models, humans’ pivotal asset for achieving (general) artificial intelligence. A recurring theme throughout the text will be the information-theoretic perspective of learning systems. In particular, we are interested in compression techniques, which have their interpretation as a “measure of intelligence”. Such interpretation is motivated by the fact that “being able to compress well, is closely related to - acting intelligently”. The intuition is that shedding redundancy from the data leads to meaningful summarizations and pattern discovery i.e generalization, one way of recognising intelligent behaviour. This thesis is a result of combining four papers which all relate to information compression. We startbypresenting anefficient approach for extracting relevant information from large datasets which enables human to consume (parse) big data. In particular, the focus in chapitre 2 is on distributed clustering recast as information compression. We then continue in chapitre 3, towards informationtheoretic causal discovery, where an optimal code length description is leveraged in distinguishing causes from effects in observational data. Although not yet entirely established, an information-theoretic interpretation is been posed as theoretical grounding and explanation of deep learning models. This brings us to chapitre 4, where we propose a generative hybrid model that exhibits explorative and flexible properties at the same time. Our generative vii model benefits from the nonparametric modeling of copulas on top of a compressed version of the training data (i.e. embedded features of an autoencoder), that makes up for an efficient, easy add-on technique, which altogether paves its way to myriad of applications. In the last chapter of the thesis, we present simple, scalable, single-model uncertainty estimates constructed by benefiting from the most informative feature representations (outputs of the last layer) of a deep model.

In conclusion, by tackling causal, generative and uncertainty-aware machine learning methods, we hope to increase the integrity and aid the adaptation of such models in any domain of application.

Climate change is pushing us to find new ways to manage energy production, distribution, and consumption. At the consumption end, smart energy meters, energy monitoring devices and applications, and renewable energy technologies such as solar photovoltaic and battery storages empower energy consumers to evolve into prosumers: the producers and consumers of energy. This research studies the prosumer role in the sustainable energy system.

The research presents two main perspectives on prosumerism; it explores both the micro-foundations and macro-level influences on the energy prosumers. The research results are displayed in the form of six articles published in international peer-reviewed journals and conferences. The first two articles make propositions about the prosumer role as part of the changing socio-technical energy and innovation system. The next two articles focus on understanding the micro-level impact on the energy prosumers and examine the producer–consumer, in particular, as a co-creator of energy-related innovations. The remaining two articles address the impact of macro-level policies on prosumers.

Theoretical contributions of the research are related to the novel research framework that combines the concepts from the socio-technical multi-level perspective, innovation studies, and policy research. Practical contributions of the study are related to the understanding of the micro-foundations of prosumer interests toward innovation co-creation activities. Practitioners benefit from evidence concerning the differences between consumers and prosumers, which may help them in better designing products and services. Policy-makers may benefit from the findings related to the policy analysis that studies policy influence on the prosumers and compares different prosumer activities with policy mixes and calls for a more holistic and systemic approach for the development of the prosumer related policies.

In the framework of a joint supervision of doctoral studies with Tampere University.

A considerable share of mathematical research has its roots in physics and has progressed to present form through abstraction and generalization. Topological dynamics for instance traces back to the study of physical systems changing over time. Its objects of enquiry are flows, which generalize a variety of real-world systems in biology, economics and engineering. Topological dynamics is concerned with the asymptotic or long term behavior of flows, answering questions like: will the system eventually return to its initial state?
A game-changing theorem by Kechris, Pestov, and Todorcevic established a connection between topological dynamics and Ramsey theory. Contrary to topological dynamics, the genesis of Ramsey theory lies in abstract mathematics. Results in this field state that one can find well-organized subsets within large disorganized sets, or find order in chaos, as they are commonly popularized. A prototypical example is that in a group of 6 people there are always at least three people which either all know each other or are all strangers.
Ramsey theory has numerous applications, from number theory – Green-Tao's famous proof that prime numbers contain arbitrarily long arithmetic progressions, for example – to theoretical computer science – decidability of constraint satisfaction problems, which have consequences in A.I. research, for instance.
In this thesis, we establish new results in projective Fraïssé theory, a field with deep implications to the study of topological dynamics of groups of homeomorphisms of compact spaces.

In the framework of a joint supervision of doctoral studies with the Università degli Studi di Torino.

The series of interactions between service providers and customers are called customer journeys. These customer journeys, today, are highly personalized, due to the new devices and technologies that are available. At the same time, new methods are required to help businesses better understand customer behavior. Business process management (or BPM) is a discipline that aims to optimize a company's processes. Process Mining is a branch that complements BPM by providing a data-driven analysis approach. In this dissertation, we investigate the ways in which process mining and BPM can help to increase businesses' comprehension of customer journeys. One of the key findings is that both the process mining framework and the XES standard for storing event logs in process mining settings are relevant for customer journeys. We show that some process mining activities can be applied as-is while other techniques need to take into account the specifics of customer journeys. In particular, we contribute by proposing new algorithms for discovering, enhancing, and exploring customer journeys. We also propose new techniques for predicting next customer interactions. Overall, we contribute by leveraging process mining know-how to improve customer journey analytics; two disciplines that were, to the best of our knowledge, never before considered together.

With cloud and mobile computing, a new category of software products emerges as mass-market information systems (IS) that addresses distributed and heterogeneous end-users. Understanding user requirements and the factors that drive user adoption are crucial for successful design of such systems. Existing IS theories and models contribute to a theoretical understanding of the adoption and use of IS in mass-markets, however they are criticized for not being able to drive actionable insights on IS design as they consider the IT artifact as a black-box. We argue that IS needs to embrace market research techniques to understand and empirically assess user preferences and perceptions in order to integrate the "voice of the customer" in a mass-market scenario. We aim at supporting the design of mass-market IS by establishing a reliable understanding of consumer’s preferences for multiple factors combing functional, non-functional and economic aspects. We apply our findings to the privacy-aware design of mass-market IS and evaluate their implications on user adoption.

Economic offshoring has been a major trend for the past 30 years. Spreading supply-chains always thinner across the globe to reach lower labor costs has shown its limits in terms of intellectual property risks, responsiveness to demand volatility and disruptions, ethical and environmental impact... Moreover, recent research has shown that local production in high-cost countries can be competitive when taking mismatch costs into account, and with an adequate product portfolio strategy.
But still, the offshoring momentum continues. Why?
In the first part of this thesis, I explore offshoring through the lens of behavioral decision-making, and take a first step toward the greater goal of identifying heuristics at play in offshoring decisions. I developed a software-based trial that puts 200 participants from various background in front of an offshoring decision focused on a mismatch costs problem: offshoring production to benefit from lower costs but facing demand uncertainty, or producing locally at a higher-cost once demand is known.
In the second part of the thesis, I present a simulation-game that I developed to help transmit these research insights to students, managers and policymakers. Indeed, my goal with this research project is to make a new body of knowledge on reshoring accessible to a wide public through an active learning approach – gamification, trial and errors, experiential learning – so that the scientific breakthroughs do not stay locked in the laboratory, and instead spread to empower practitioners by improving their “heuristic toolbox”.

Software problems do not only induce high financial loss, but also sometimes induce human loss. Those problems are due to the presence of software bugs, failures, errors, and defects in software systems. These software anomalies, and in particular the software defects, have a huge impact not only on business activities but also on the cost of developing and maintaining these software systems. In order to identify their sources, particularly the ones causing severe impacts on the systems’ operations, we conducted two case studies. We analyzed software defects of two systems over a period of a year and a half. We classified these software defects, according to their trigger factors and according to their severity impact. Conducting these studies led us to propose “the origins of severe software defects method” order to identify trigger factors that cause severe software defects on a given evolving system. We also found that the group of technology trigger factors causes more severe defects than the other groups of trigger factors for this type of systems.

This thesis studies the value of responsiveness for a manufacturer. In practical terms, responsiveness allows a manufacturer to wait until actual customer demand can be observed. The benefits of this come from minimized waste, when less unwanted goods are produced, and increased sales as more of the actual demand can be met. If a manufacturer uses responsiveness to provide services, and for example, customization of products to its customers, responsiveness can be a strategic advantage in competition. This work contributes to the understanding of when the higher cost of responsiveness can be justified and manufacturer should invest in local capacity instead of low-cost offshore manufacturing.

The first essay, investigates an approach called the Volatility Portfolio and Option-based Costing. This approach suggests building a balanced portfolio of products with high and low time-sensitivity to maximize the benefits from responsiveness. This advice is contrary to the common intuitive solution. With four applications in cases from different industries, we show how the approach delivers value and how to move from theory to practice.

It has been shown that innovation follows manufacturing. Using company cases, I build a hypothesis of how responsiveness leads to higher innovation, because of local problem-solving and customer-driven innovations. The contribution from the essay, is that companies should think about learning and innovation as they make their production decisions.

Currently used models can systematically underestimate the value of lead time reduction. This can be significant for products with short sales period and a clearance price that varies in the share of unsold inventory. Third essay, demonstrates reasons for this underestimation and provides tools to fix it.

This thesis presents a behavioral economics contribution to the security of information systems. It focuses on security information sharing (SIS) between operators of critical infrastructures, such as systemic banks, power grids, or telecommunications. SIS is an activity by which these operators exchange cybersecurity-relevant information, for instance on vulnerabilities, malwares, data breaches, etc. Such information sharing is a low-cost and efficient way by which the defenders of such infrastructures can enhance cybersecurity. However, despite this advantage, economic (dis)incentives, such as the free-rider problem, often reduce the extent to which SIS is actually used in practice. This thesis responds to this problem with three published articles.

The first article sets out a theoretical framework that proposes an association between human behavior and SIS outcomes. The second article further develops and empirically tests this proposed association, using data from a self-developed psychometric survey among all participants of the Swiss Reporting and Analysis Centre for Information Assurance (MELANI). SIS is measured by a dual approach (intensity and frequency), and hypotheses on five salient factors that are likely associated with SIS outcomes (attitude, reciprocity, executional cost, reputation, trust) are tested. In the third article, policy recommendations are presented in order to reduce executional costs, which is found to be significantly and negatively associated with SIS. In conclusion, this thesis proposes multiple scientific and practical contributions. It extends the scientific literature on the economics of cybersecurity with three contributions on the human factor in SIS. In addition, regulators will find many recommendations, particularly in the area of governance, to support SIS at the legislative level. This thesis also offers many avenues for practitioners to improve the efficiency of SIS, particularly within Information Sharing and Analysis Centers (ISACs) in charge of producing Cyber Threat Intelligence in order to anticipate and prevent cyberrisks.

It is commonly acknowledged that business model innovation carries enormous opportunities for incumbent organizations, especially when driven by digital transformation. However, less is known and discussed about the challenges for off-line born organizations – i.e. established before the diffusion of the Internet - which attempt to tackle this journey of change. In this context, thanks to my research setting, based on the collaboration between the BISA team at UNIL and SAP AG, I contribute to the business model domain with two research streams.

First, I address the process of business model management, analyzing phases that go beyond business model design. I observe this process in practice, complementing the predominantly conceptual literature. I contribute to the research by identifying two approaches to business model management: a deterministic and waterfall approach, characterized by a high level of certainty and confidence by the management team; and a discovery-driven approach, in which numerous design and evaluation iterations are performed before business model implementation.

Second, I study the design of business models for connected products. Phenomena like internet of things and smart cities require a complex network of actors in which organizations, individuals, and objects exchange value. Existing business model representations are not fully capable of describing such networks, having rather generic elements and components. Therefore, I take a first step towards new means of representation, proposing a taxonomy of design elements to represent business models for cyber-physical systems, the combination of physical and computational processes at the foundation of connected products. The main contribution of this research is a specific set of actors’ roles, the type of value they exchange and perceive, as well as their dominance in the network.

In this doctoral dissertation, I relate three studies performed to address the challenge of a visual inquiry tool for identity communication in the context of startups and small and medium enterprises. The challenge being: how to develop a visual inquiry tool (a tool on which a team of stakeholders with different backgrounds could try and solve their challenge in a designerly way) especially tailored to help them tackle the issue of communicating a coherent brand identity to all their different stakeholders. These three chapters (or studies) have been developed within a design science paradigm of research, which allows to develop knowledge through both theoretical and in the form of artefacts to tackle a practical problem. The main contributions of this dissertation are: 1) a brand identity ontology based on an extensive literature review, which addresses the semantic issues found in the brand identity literature and gives us the opportunity to explore and redefine the concept in terms of a conceptual model and 2) an identity communication map, this is derived from the ontology but is this time directly aimed at practitioners. It addresses the challenge of creating a coherent and structured identity communication especially in the context of startup and SMEs. And lastly, 3) by analyzing existing visual inquiry tools, we derived a design theory for managing any business challenge in a designerly way. This last contribution aims at supporting future designers and researchers when developing such artefacts. The view proposed in this thesis is  highly interdisciplinary, but focuses mainly on design and proposes to adopt a new approach when solving management problems.

In this thesis, I investigate two specific subjects in data science, namely demand forecasting and causality inference, dividing this thesis in two main parts. The first part aims at improving demand forecasting accuracy that impacts supply chain performance. It consists of three articles aiming at studying how to enhance demand forecasting accuracy using pertinent data (e.g. operational transaction data, weather data, socio-economic data, etc.). Each article explores a new statistical approach on the supply chain optimization through demand forecasting accuracy. We found that considered pertinent data have a significant impact on demand forecasting accuracy with reductions in percentage errors up to 48%. These results can be used to justify and motivate the integration of pertinent data in the decision making process in order to better anticipate demand volumes and reduce costs due to excess inventory or stock shortages.

The goal of the second part is to infer the causal relationship in the case of non-linearity and heteroscedasticity, meaning when the variance is not constant. We provide a bivariate multiplicative noise model (Causal Heteroscedastic Model, CHM) that we extend to the multiplicative case. This two-steps method infers the intrinsic causal mechanism ; it consists of applying a causal additive model on the BAMLSS (bayesian additive model for location, scale and shape) fitted values of the estimated parameters. The simulation study provides an accuracy of 0.97 on average, and the application of CHM on financial indices shows an un-lagged causal effect of the shares on the index they compose.

Digitalization and changing customer demands prompt many organizations to scrutinize their existing and develop new business models (BMs). Against this backdrop, the BM concept has flourished as a key theoretical and practical device for modeling, analyzing, describing, and designing how firms create, deliver, and capture value. Despite the rapidly growing body of literature on BMs, prior research provides very little insight into the challenges that arise when we shift focus from design to BM management as a holistic process, and from one to multiple BMs per firm. Based on qualitative and exploratory research designs, this cumulative dissertation examines three interrelated aspects in the realm of BM management in the multi-BM firm: BM management processes, BM tools, and BM portfolios.

First, it determines activities, phases and characteristics of BM management processes, especially in comparison to the theoretical and idealistic processes proposed in earlier literature. It uncovers two distinct kinds of BM management processes: discovery-driven and deterministic. Second, it analyzes social practices and knowledge boundaries between communities of practice in BM innovation, providing implications for BM tool design. Third, it conceptualizes BM portfolio management and provides a classification of BM combinations based on their sources of value creation.

Taken together, this dissertation analyzes rich empirical data to cast light on the actual mechanisms and social practices that unfold in organizations when managing multiple BMs. Results point at inherent variety and dynamics in the ways how BMs are managed along their lifecycle, vis-à-vis complex social processes. It hereby contributes a complementary perspective to prevalent scientific discussions on what BMs are, by focusing on BMs as something that people do.

Realistic models of human mobility are critical for modern day applications, specifically for recommendation systems, resource planning and process optimization domains. Given the rapid proliferation of mobile devices equipped with Internet connectivity and GPS functionality today, aggregating large sums of individual geolocation data is feasible. The thesis focusses on methodologies to facilitate data-driven mobility modeling by drawing parallels between the inherent nature of mobility trajectories, statistical physics and information theory. On the applied side, the thesis contributions lie in leveraging the formulated mobility models to construct prediction workflows by adopting a privacy-by-design perspective. This enables end users to derive utility from location-based services while preserving their location privacy. Finally, the thesis presents several approaches to generate large-scale synthetic mobility datasets by applying machine learning approaches to facilitate experimental reproducibility.

Cross-boundary teams (those consisting of members across functions and organizations) have been considered in the past two decades as the most effective strategy for organizations to undertake complex and innovative projects. While these teams are well-equipped to undertake such projects as they can tap on their diverse set of knowledge and skills, these differences also increase the costs and difficulties of collaborating. In this dissertation, I relate the

design science research project I undertook to address three challenges that such teams encounter frequently, i.e. the challenge of coordinating, cooperating, and solving wicked problems.
The particularity of this dissertation is that I integrate works in information systems, psycholinguistics and sociology to develop both descriptive and prescriptive knowledge on cross-boundary challenges. The prescriptive knowledge consists (1) of two artifacts (i.e., the Coopilot App and the Team Alignment Map) for the challenges of coordination and cooperation, and (2) a design theory that helps designers develop visual inquiry tools. The descriptive knowledge consists of (1) a process model for team coordination through conversation and (2) a conceptual model that informs how team members can overcome the three challenges by entering a process of joint inquiry.

Overall, I argue that for future research to contribute with prescriptive guidance for the variety of challenges cross-boundary teams can encounter, cross-boundary teamwork should be conceived of as a process of joint inquiry. Through this dissertation I also provide an illustration of how design science researchers can contribute to the lack of prescriptive and actionable knowledge for cross-boundary teamwork.

Abstract

In the framework of a joint supervision of doctoral studies with the Università degli Studi di Torino

abstract EN.pdf  (84 Ko)

Every day we make tens of guesses. How likely is Macron to win the second round of the French presidential elections of 2017? Is Brazil or Germany going to win the football world cup of 2018? Is Roger Federer or Grigor Dimitrov going to win the finals in the Rotterdam Open? This thesis investigates and models how people make such guesses. The first chapter addresses the first of those questions: How likely is something to happen? It demonstrates how we can extend an already existing theory of how we judge how likely something is so that it makes more refined predictions about our behavior. The rest of the thesis focuses on the other two questions: Which of two items is better on some dimension? The second chapter outlines how we can predict how long people will take to make such guesses and what brain regions will be active while they are making those guesses. The last chapter demonstrates that there are systematic differences between how people guess when faced with real-world questions and with typical, artificial, experimental questions. One major difference is that in the real world some objects are more familiar than other objects and people are likely to guess that the more familiar object is better. The results of this thesis help us to better understand how we make decisions and can be used by, for example, software application designers or marketing specialists to distribute information in a way that reduces the effort that we exert when faced with a decision.

Today, to find meeting points or information on public transportation, we frequently use our mobile devices and, more specifically, the location-based services installed on them. These applications are extremely convenient to use and help us on a daily basis. However, we sacrifice, sometimes without realizing it, our privacy by sharing our locations with location-based services, hence, giving private information about ourselves to companies that own these services. Indeed, information can be easily extracted from our location history, in particular, our frequently visited places, our hobbies, even our identity. This is more critical when these services, in order to create new content, build mobility-prediction models about our lives. These companies collect a large amount of personal data that can be used for commercial or malicious purposes. Consequently, it is crucial to create new algorithms and architectures that preserve our location privacy when we use location-based services. This thesis focuses on the issues exposed above. This is particularly relevant today when we know that the new European General Data Protection Regulation (EU GDPR), which aims at preserving the privacy of individuals, came into effect on May, 25^th 2018, and this implies that European and some Swiss companies must be compliant with it.

In recent years, cyber operations and malicious cyber activities have become common means of achieving strategic national interests. Their increasingly disruptive effects, which have destabilized international peace and security and fueled geopolitical instability, have catapulted cyber risks to the national and international security agenda. This dissertation explores Swiss foreign policy as an instrument conducive to international cyber stability. However, while cyberspace has developed into a distinct realm of interstate relations, I argue that states’ behavior in that domain is embedded in the existing international order, which is conducive to international peace and stability. My overall conclusion is that, for the time being, no new rules, instruments or policy responses are needed to delineate acceptable interstate behavior. This standpoint reflects the Swiss foreign policy between 2012 and 2017 to the extent that the chosen policy instruments and diplomatic processes in the cyber realm do not differ significantly from those in other Swiss policy areas. I illustrate it via three qualitative case studies, highlighting not only the multilateral venues of Switzerland’s engagement in the cyber realm (i.e. the UN and the OSCE) but also various norms and measures to build both confidence and cyber security capacity.

Random graphs theory has been an important tool to model and solve problems related to real world networks. Although those problems come from very different fields, such as for example social networks, electrical power grids, and Internet network, they share an important common feature such as a very large number of elements. Due to their large and intricate structures those problems have been first studied in terms of their elements (nodes) and connections between those elements (edges).

Very often the problems are so complicated that a complete description of the dynamics happening in the whole network is impossible. Hence there has been given a lot of attention to the local properties of the network such as how many nodes are involved in a process or how to estimate the probability that the elements of the network will interact with each other in order to produce a certain result.

In this thesis we will focus the attention on a particular category of neural network, i.e., a network which mimics the dynamics and the connectivity of neurons in the brain. The nodes of the network are representing neurons, while the edges connecting them are potential synaptic connections. We propose and analyse a random graph model which may predict synaptic formation of a network and formation of connected clusters which communicate with each other. In particular, in the resulted networks the probability of connections depends on the distance.

In the framework of a joint supervision of doctoral studies with the University of Lund.

During the past four decades, due to miniaturization computing devices have become ubiquitous and pervasive. Today, the number of objects connected to the Internet is increasing at a rapid pace and this trend does not seem to be slowing down. These objects, which can be smartphones, vehicles, or any kind of sensors, generate large amounts of data that are almost always associated with a spatio-temporal context. The amount of this data is often so large that their processing requires the creation of a distributed system, which involves the cooperation of several computers. The ability to process these data is important for society. For example: the data collected during car journeys makes it possible to avoid traffic jams, to know about the need to organize a carpool, or to plan the maintenance interventions to be carried out on the road network. The application domains are therefore numerous, as are the problems associated with them. The articles that make up this thesis deal with systems that share two key characteristics: a spatio-temporal context and a decentralized architecture. In addition, the systems described in these articles revolve around three temporal perspectives: the present, the past, and the future. Systems associated with the present perspective enable a very large number of connected objects to communicate in near real-time, according to a spatial context. Our contributions in this area enable this type of decentralized system to be scaled-out on commodity hardware, i.e., to adapt as the volume of data that arrives in the system increases. Systems associated with the past perspective, often referred to as trajectory indexes, are intended for the access to the large volume of spatio-temporal data collected by connected objects. Our contributions in this area makes it possible to handle particularly dense trajectory datasets, a problem that has not been addressed previously. Finally, systems associated with the future perspective rely on past trajectories to predict the trajectories that the connected objects will follow. Our contributions predict the trajectories followed by connected objects with a previously unmet granularity. Although involving different domains, these contributions open the possibility of being able to deal with these problems more generically in the near future.

Information technologies (IT) have had a massive impact on the capacities of organizations to access and treat information, which have eventually increased their productivity. They have become so integrated in routines that without them, organizations are unable to operate. As an example, in August 2016, Delta Airlines was obliged to cancel almost 2,000 flights because its central system broke down. With the growing capacity of IT, business applications (e.g., enterprise systems) have been supporting increasingly complicated and individual tasks. However, these applications are often chosen based on an organization’s objectives with little consideration for individual needs. They push standardized routines that ask employees to change theirs. This results in a large part of employees being unsatisfied with the way business applications support their activities. With the increasing capacities of mobile applications, many employees have shifted from desktop to mobile applications. Because mobile devices are personal, their applications should be designed to adapt to individual work patterns. 
However, despite their popularity and efficiency, very few studies have investigated the designs of mobile applications and their uses in organizational contexts. This dissertation addresses this gap through three interrelated research streams: Research stream 1 investigates the interplay between individual routines and mobile apps as IT artifact. To do so, it looks into the roles that mobile apps play in supporting the ostensive and performative aspects of individual routines and the underlying design of mobile apps’ user interfaces based on two field studies: customer interactive support and routine patient care. Research stream 2 looks into the capacities of mobile checklists, i.e. checklists that are executed on mobile devices, to codify and execute routines. Checklists are a very efficient structure to support individual routines, as described in existing literature and also in the two mobile apps analyzed in the research stream 1. Given the roles of checklists and their frequent uses in mobile applications, I investigate how organizational knowledge is codified and adapted to different contexts as well as how tasks are documented and validated. Research stream 3 seeks to analyze the structures and the components of individual routines in order to describe, assess and improve them. It intends to understand the extent to which activity patterns are structured vs. unstructured and the uses of IT artifacts in these patterns. Thus, it investigates the use of maturity models and process mining to support organizations in analyzing and improving their routines. 
To conclude this dissertation, I discuss the application of my contributions in view of an ongoing project involving the use of smart glasses to support individual routines as well as the links between this dissertation and existing research in human-computer interaction. 

Over the past decades the electricity supply has been reliable in Switzerland. However, it is uncertain how the electricity supply will evolve in the long-term given the potential changes in the generation-mix in Switzerland, resulting from the nuclear phase-out and the increasing share of PV. The objective of this research is to elaborate on the concept of security of supply in the electricity sector (SoES), and to analyse in particular the case of Switzerland.

We develop a system dynamics model to analyse the impact of these changes on SoES in the long-term. Our results show that with the current regulatory framework, the only investments committed to are those assumed for PV and wind energy until 2035. Consequently, generation adequacy deteriorates progressively and the country becomes a net importer. Given the recent large investments in pumped-storage power plants (PSP), we also analyse how the changes in the Swiss electricity market threaten their profitability. We develop an algorithm to simulate PSP operation and integrate it into our model. Although the changes in the generation-mix lead to higher price differences, the drop of available cheap energy lead to low arbitrage opportunities. As current electricity systems are very complex, the elements in our model are not the only ones affecting the SoES.. Based on a literature review, we develop a framework comprising twelve dimensions, which cover all aspects of long-term SoES. We provide at least one metric for each dimension.

Our overall conclusion is that the security of supply is threatened in Switzerland. In particular, the nuclear phase-out, whatever its timing, will have major effects on prices and on the country’s self-sufficiency. Our framework can be used to monitor the electricity market over time in order to provide insights about the expected evolution of all the aspects of SoES and provide guidance for action.

Sensory information processing is a key process in the brain because it involves many sensory inputs. Some of them are relevant and should induce a motor or cognitive response. In addition, many irrelevant stimuli reach sensory pathway and should be ignored. Synaptic plasticity in the central nervous system is a general process that enhances or decreases sensory responses according to the temporal pattern of stimuli. My main aim is to study synaptic plasticity in the somatosensory pathway, mainly in the thalamo-cortical loop. Sensory information from rodent whiskers is sent from the whisker follicle to the contralateral area of the thalamus and from the thalamus to the barrel cortex (BC). In this Doctoral Thesis we performed extracellular in vivo recordings in the BC and thalamus of urethane anesthetized rats and mice in order to unravel the mechanisms of synaptic plasticity and sensory processing. We observed that repetitive stimulation at frequencies at which the animal explores the environment induced Jong-term potentiation (LTP). In addition, low frequency stimulation could induce LTP or long-term depression (LTD) depending on the intracellular Ca²⁺ concentration during the stimulation time period. This long-term plasticity depended on NMDA receptors activation and the activation of muscarinic and nicotinic cholinergic receptors. Through an optogenetic study we showed that the basal forebrain (BF), the main source of acetylcholine (Ach) to the neocortex, sent its projections in an organized way. Consequently, the Ach-depending facilitation of cortical responses occurs in a very specific manner. We also found that the postero-medial thalamic nucleus (POM) regulated BC whisker responses through GABAergic (ɣ-aminobutyric-acid: GABA) neurons located in upper cortical layers.

Thesis in joint-supervision with the Universidad Autónoma de Madrid

Full text of the thesis avaiable on Serval: https://serval.unil.ch/notice/serval:BIB_E66BE6750A09

Predictions that information technology (IT) will become a dominant driver in patient care delivery continue to proliferate. While IT’s potential benefits in healthcare are manifold, past research has shown that the digitalization of medicine remains more of a promise than a reality. Given the current limitations of IT in healthcare, in this dissertation, I argue that we need prescriptive design knowledge on how IT artifacts ought to be to function in patient care delivery.

Investigating two dominant IT artifacts in healthcare, electronic health record (EHR) systems and mobile medical apps, this dissertation unpacks the ‘black box’ of IT artifacts and sheds light on the design and the effective use of IT artifacts in routine patient care. The dissertation comprises three interrelated research streams, each taking a specific angle to study the aforementioned aspects. Research stream 1 provides a conceptual framework on the interdependencies between routines in patient care and EHR systems and devises two strategies to objectively assess and improve the effective use of EHR systems. Rather than studying design and use of EHR systems separate from each other, our framework suggests combine the two. Research stream 2 investigates routines at the individual level and describes the affordances of mobile apps to accommodate differences in EHR system use among individual physicians. Research stream 3 centers around the patient and studies the design of medical apps to provide a way for patients to self-diagnose their acute symptoms and to enhance the monitoring of an illness. This research stream presents design principles that effective medical apps should possess in order to engage the patient in the delivery of care. The theoretical contributions can be classified as mid-range theories and inform design practice by being specific about both users (i.e. patients and physicians) and IT artifacts (i.e. EHR systems and medical apps).

Motivated by both economic and environmental reasons, there has been a boost in the number of studies attempting to improve the energy efficiency of data centers (DCs). However, after analyzing the literature related to the improvement of the efficiency of DCs, we spotted several gaps that are assessed in this thesis.

In the first paper, we study how operations management principles apply to DCs running scientific jobs. In particular, we test Little’s law, the law of variability and the law of utilization by using data from three major scientific data centers. Results show that both, Little’s law and the law of utilization, hold, while the law of variability does not. These findings give insights into how DC operations can be improved by applying operations management principles to DCs.

The second paper reviews the most commonly used performance indicators in DCs. We analyze these indicators and find several drawbacks. To overcome these drawbacks, we develop a new performance indicator that has all the characteristics that a normal performance indicator should have according to the literature. The proposed indicator is evaluated using data from three DCs and by controlled lab-tests.

In the third paper, a two-step method to model and forecast the Load-at-Risk of computing systems is proposed. Data from a Finnish company’s computing system is used to validate the method developed. Results show that the two-step method successfully models and forecasts the Load-at-Risk of computing systems. This provides operators of computing systems with better information regarding how high the workload of the system could be with a significance level.

To sum up, this research provides means to assist the managers and operators of computing systems to improve the efficiency at which their computing resources are operated.

This thesis is devoted to the study of non-Borel ∆12 pointclasses of the Baire space, using reductions by continuous functions. This work is divided in three main parts. In the first one, we generalise results obtained by Duparc and Louveau to provide a complete description of the Wadge hierarchy of the class of increasing di˙erences of coanalytic sets, under some determinacy hypothesis. In a second part, we study some ∆12 pointclasses above the class of increasing di˙erences of coanalytic sets, and give a fragment of the Wadge hierarchy for those classes. Finally, we apply our results and techniques to theoretical computer science and more precisely to the study of regular tree languages, that is sets of labeled binary trees that are recognized by tree automata.

Thesis in co-supervision with the Université Paris-Diderot (Paris 7)

In this thesis, we develop tools to study the influence of predictors on multivariate distributions. We tackle the issue of conditional dependence modeling using generalized additive models, a natural extension of linear and generalized linear models allowing for smooth functions of the covariates. Compared to existing methods, the framework that we develop has two main advantages. First, it is completely flexible, in the sense that the dependence structure can vary with an arbitrary set of covariates in a parametric, nonparametric or semiparametric way. Second, it is both quick and numerically stable, which means that it is suitable for exploratory data analysis and stepwise model building. Starting from the bivariate case, we extend our framework to pair-copula constructions, and open new possibilities for further applied and methodological work. Our regression-like theory of the dependence, being built on conditional copulas and generalized additive models, is at the same time theoretically sound and practically useful.

The ubiquity of mobile devices and particularly smartphones has caused the emergence of a new trend of distributed applications known as Proximity-Based Mobile (PBM) applications. These applications enable a user to interact with others in a defined range and for a certain time duration for different purposes such as social networking, dating, gaming and driving. The goal of this thesis is to introduce a set of programming abstractions and algorithms that can be used for building PBM applications in a category of mobile networks, called mobile ad hoc networks (MANETs). In fact, the characteristics of MANETs make them a promising technology to enable PBM applications. However, the existing abstractions and algorithms in the literature of MANETs are not fully adequate for building PBM applications. Thus, in this thesis we define proximity-based durable broadcast and proximity-based neighbor detection as the main requirements of PBM applications. Then, in each part of the thesis, we introduce abstractions and algorithms which address one of these requirements.