付記する学位プログラム名: デザイン学大学院連携プログラム / 京都大学 / 新制・課程博士 / 博士(情報学) / 甲第25418号 / 情博第856号 / 新制||情||143(附属図書館) / 京都大学大学院情報学研究科知能情報学専攻 / (主査)特定教授 黒橋 禎夫, 教授 河原 達也, 教授 楠見 孝 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

De plus en plus d’organisations scientifiques contemporaines intègrent dans leur processus des foules de participants assignés à des tâches variées, souvent appelés projets de science citoyenne. Ces foules sont une opportunité dans un contexte lié à une avalanche de données massives qui met les structures scientifiques face à leurs limites en terme de ressources et en capacités. Mais ces nouvelles formes de coopération sont déstabilisées par leur nature même dès lors que les tâches déléguées à la foule demandent une certaine inventivité - résoudre des problèmes, formuler des hypothèses scientifiques - et que ces projets sont amenés à être répétés dans l’organisation. A partir de deux études expérimentales basées sur une modélisation originale, cette thèse étudie les mécanismes gestionnaires à mettre en place pour assurer la performance des projets délégués à la foule. Nous montrons que la performance est liée à la gestion de deux types de capitalisation : une capitalisation croisée (chaque participant peut réutiliser les travaux des autres participants) ; une capitalisation séquentielle (capitalisation par les participants puis par les organisateurs). Par ailleurs cette recherche met en avant la figure d’une nouvelle figure managériale pour supporter la capitalisation, le « gestionnaire des foules inventives », indispensable pour le succès des projets. / A growing number of contemporary scientific organizations collaborate with crowds for diverse tasks of the scientific process. These collaborations are often designed as citizen science projects. The collaboration is an opportunity for scientific structures in a context of massive data deluge which lead organizations to face limits in terms of resources and capabilities. However, in such new forms of cooperation a major crisis is caused when tasks delegated to the crowd require a certain inventiveness - solving problems, formulating scientific hypotheses - and when these projects have to be repeated in the organization. From two experimental studies based on an original modeling, this thesis studies the management mechanisms needed to ensure the performance of projects delegated to the crowd. We show that the performance is linked to the management of two types of capitalization: a cross-capitalization (each participant can reuse the work of the other participants); a sequential capitalization (capitalization by the participants then by the organizers). In addition, this research highlights the figure of a new managerial figure to support the capitalization, the "manager of inventive crowds", essential for the success of the projects.

[pt] Este trabalho propõe um modelo de otimização robusta de pior caso orientado por dados aplicado na seleção de um portfólio de títulos de renda fixa. A gestão das carteiras implica na tomada de decisões financeiras e no gerenciamento do risco através da seleção ótima de ativos com base nos retornos esperados. Como estes são variáveis aleatórias incertas foi incluído um conjunto definido de incertezas estimadas diretamente no processo de otimização, chamados de cenários. Foi usado o modelo de ajuste de curvas Nelson e Siegel para construir as estruturas a termo das taxas de juros empregadas na precificação dos títulos, um ativo livre de risco e alguns ativos com risco de maturidades diferentes. Os títulos prefixados são marcados a mercado porque estão sendo negociados antes do prazo de vencimento. A implementação ocorreu pela simulação computacional usando dados de mercado e dados estimados que alimentaram o modelo.Com a modelagem de otimização robusta foram realizados diferentes testes como: analisar a sensibilidade do modelo frente às variações dos parâmetros verificando seus resultados e a utilização de um horizonte de janela rolante para simular o comportamento ao longo do tempo. Obtidas as composições ótimas das carteiras, foi feito o backtesting para avaliar o comportamento das alocações com o retorno real e também a comparação com o desempenho de umbenchmark. Os resultados dos testes mostraram a adequação do modelo da curva de juros e bons resultados de alocação do portfólio robusto, que apresentaram confiabilidade até em períodos de crise. / [en] This paper proposes a data-driven worst case robust optimization model applied in the selection of a portfolio of fixed income securities. The portfolio management implies in financial decision-making and risk management through the selection of optimal assets based on expected returns. As these are uncertain random variables, was included a defined set of estimated uncertainties directly in the optimization process, called scenarios. The Nelson and Siegel curve fitting model was used to construct the term structure of the interest rates employed in the pricing of securities, a risk-free asset and some risky assets of different maturities. The fixed-rate securities are marked to market because they are being traded before the maturity date. The implementation took place through computational simulation using market data and estimated data that fed the model. With robust optimization modeling were done different tests such as: analyze the sensitivity of the model to the variations of the parameters checking the results and the use of a rolling horizon scheme to simulate behavior over time. Once the optimal portfolio composition was obtained, the backtesting was done to evaluate the behavior of the allocations with the real return and also the comparison with the performance of a benchmark. The results of the tests showed the adequacy of the interest curve model and good allocation results of the robust portfolio, which presented reliability even in times of crisis.

Various research has shown the potential and robustness of deep learning-based approaches to synthesise novel motions of 3D characters in virtual environments, such as video games and films. The models are trained with the motion data that is bound to the respective character skeleton (rig). It inflicts a limitation on the scalability and the applicability of the models since they can only learn motions from one particular rig (domain) and produce motions in that domain only. Transfer learning techniques can be used to overcome this issue and allow the models to better adapt to other domains with limited data. This work presents a study of three transfer learning techniques for the proposed Objective-driven motion generation model (OMG), which is a model for procedurally generating animations conditioned on positional and rotational objectives. Three transfer learning approaches for achieving rig-agnostic encoding (RAE) are proposed and experimented with: Feature encoding (FE), Feature clustering (FC) and Feature selection (FS), to improve the learning of the model on new domains with limited data. All three approaches demonstrate significant improvement in both the performance and the visual quality of the generated animations, when compared to the vanilla performance. The empirical results indicate that the FE and the FC approaches yield better transferring quality than the FS approach. It is inconclusive which of them performs better, but the FE approach is more computationally efficient, which makes it the more favourable choice for real-time applications. / Många studier har visat potentialen och robustheten av djupinlärningbaserade modeller för syntetisering av nya rörelse för 3D karaktärer i virtuell miljö, som datorspel och filmer. Modellerna är tränade med rörelse data som är bunden till de respektive karaktärskeletten (rig). Det begränsar skalbarheten och tillämpningsmöjligheten av modellerna, eftersom de bara kan lära sig av data från en specifik rig (domän) och därmed bara kan generera animationer i den domänen. Kunskapsöverföringsteknik (transfer learning techniques) kan användas för att överkomma denna begränsning och underlättar anpassningen av modeller på nya domäner med begränsade data. I denna avhandling presenteras en studie av tre kunskapsöverföringsmetoder för den föreslagna måldriven animationgenereringsnätverk (OMG), som är ett neural nätverk-baserad modell för att procedurellt generera animationer baserade på positionsmål och rotationsmål. Tre metoder för att uppnå rig-agnostisk kodning är presenterade och experimenterade: Feature encoding (FE), Feature clustering (FC) and Feature selection (FS), för att förbättra modellens lärande på nya domäner med begränsade data. All tre metoderna visar signifikant förbättring på både prestandan och den visuella kvaliteten av de skapade animationerna, i jämförelse med den vanilla prestandan. De empiriska resultaten indikerar att både FE och FC metoderna ger bättre överföringskvalitet än FS metoden. Det går inte att avgöra vilken av de presterar bättre, men FE metoden är mer beräkningseffektiv, vilket är fördelaktigt för real-time applikationer.

Nowadays, becoming data-driven is the vision of almost all organizations. However, achieving this vision is not as easy as it may look like and there are many factors that affect, enable, support and sustain the data-driven ecosystem in an organization. Among these factors, this study focuses on data democracy which can be defined as the intra-organizational open data that aims to empower the employees getting faster and easier access to data in order to benefit from the business insight they need without the interfere of external help.  In the existing literature, while the importance of becoming data-driven has been widely discussed, when it comes to data democracy within organizations, there is a noticeable gap. As a result, this master’s thesis aims to justify the importance and role of the data democracy in becoming a data-driven organization, focusing on the case of a Swedish bank. Additionally, it intends to provide extra investigation on the role of data analytics tools in achieving data democracy.  The results of the study show that there is a strong connection between the benefits of the empowering different actors of the organization with the needed data knowledge, and the speeding up of the data-driven transformation journey. Based on the study, shared data and the availability of data to a larger number of stakeholders inside an organization result into a better understanding of different aspects of the problems, simplify the data-driven decision making and make the organization more data-driven. In the process of becoming data-driven, the organizations should provide the analytics tools not only to the data specialists but even to the non-data technical people. And by offering the needed support, training and collaboration possibilities between the two groups of employees (data specialists and non-data specialists), it should be attempted to enable the second group to extract the insight from the data, independently from the help of the data scientists.  An organization can succeed in the path of becoming data-driven when they invest on the reusable capabilities of its employees, by discovering the data science skills across various departments and turning their domain experts into citizen data scientists of the organization.

In an increasingly digital world, companies need to be at the forefront of development in order to gain market share and be competitive. Therefore this qualitative case study intends to investigate how Business intelligence should be implemented to increase the technology acceptance among the employees. Furthermore, it is investigated how data-driven a consulting company in Sweden is. An important factor when Business intelligence is about to be implemented and applied are the employees and the company culture. It is important to normalize the collection of data, in order to create a data culture where high quality data is collected. What is more, Business intelligence is strongly dependent on the data collected maintaining a high data quality, in order to be able to create relevant reports and thereby be able to support various decision-making. When implementing a new platform, many employees are affected. This means that the platform can generate both positive and negative reactions. / I en allt mer digitaliserad värld krävs det att företag ligger långt fram i utvecklingen för att kunna ta marknadsandelar och vara konkurrenskraftiga. Därmed ämnar denna kvalitativa fallstudie att undersöka hur Business intelligence kan implementeras för att öka acceptansen för IT-stöd hos de anställda, samt hur datadrivet ett svenskt konsultföretag är. En viktig faktor när Business intelligence ska implementeras och appliceras är de anställda och den kultur som företaget har. Det är viktigt att normalisera insamlandet av data, för att i förlängningen skapa en datakultur där data med hög kvalité samlas in. Vidare är Business intelligence starkt beroende av att den data som samlas in håller en hög datakvalité, för att kunna skapa relevanta rapporter och därigenom kunna ge förslag inför olika beslutsfattande. Vid en implementering av en ny plattform, är det många anställda som berörs. Det betyder att plattformen kan generera såväl positiva som negativa reaktioner.

The digitalization of our society and the creation of data intense industries are transforming how industrial sales can be made. Large volumes of data are generated when businesses and people use digital products and services which are available in the modern world. Some of this data describes the digital products and services when they are in use, i.e., it is in-service data. Furthermore, data has during the last decade been seen as an asset which can improve decision-making and has made sales activities become increasingly customer specific. The purpose of this study was to explore how knowledge from in-service data can serve B2B selling. To realize this purpose the following three research questions were answered by conducting a single case study of a large company in the information and communications technology (ICT) industry. (RQ1) How does a company in a data intense industry use knowledge from in-service data in the B2B sales process? (RQ2) What opportunities does knowledge from in-service data create in the B2B sales process? (RQ3) What challenges hinder a company from using knowledge from in-service data in the B2B sales process? RQ1: This study has concluded that, in the context of a data intense industry, throughout the steps in the B2B sales process, knowledge from in-service data is actively used by the sales team, however, to varying degrees. In-service data is used in six categories of sales activities: (1) to understand the customer in terms of their technical and strategical needs, which enables lead generation and cross-selling, (2) to make information from in-service data available through data collection, storage, and analyses, (3) to nurture the relationship between buyer and seller by creating understanding, trust and satisfactory offers to the customer, (4) to present solutions with convincing arguments, (5) to solve problems and satisfy the customer’s needs, and (6) to provide post-sale value-adding services. Moreover, three general resources which are used in the activities were identified: An audit report which presents the information of the data, a plan which presents strategic expansions of the solution, and simulations of the solution. Furthermore, four general actors who are performing the activities were identified: the Key Account Manager (KAM) who is responsible for conducting the sales interactions with the customer, the sales team, and the presales team who both support the KAM, and the customer. In addition to the general resources and actors, companies may use step-specific resources and actors. RQ2: Four categories of opportunities were identified: knowledge from in-service data (1) assists KAMs in discovering customer needs, (2) guides the KAM in creating better customer specific solutions, (3) helps the KAM move the sale faster through the sales process, and (4) assists the company in becoming a true partner who provides strategic services, rather than acting as a supplier. RQ3: Finally, four categories of challenges were identified: (1) organizational, (2) technological, (3) cultural, and (4) legal & security. Out of these, obtaining access to the data was identified as the greatest challenge to use in-service data. The opportunities and the challenge to access data are deemed to be general for companies in data intense industries, while the other challenges are depending on the structure, size, and culture of the individual company. The findings of this study contribute to a general understanding of how companies in data intense industries may use knowledge from in-service data, what opportunities this data create for their B2B sales process, and which challenges they face when they pursue activities which use the knowledge from in-service data. To conclude, in-service data serves B2B selling especially as a source of customer knowledge. It is used by salespeople to understand the customer in terms of its technical and strategical needs and salespeople use this knowledge to conduct various customer-oriented sales activities. In-service data creates several opportunities in B2B sales. However, several challenges must be overcome to seize the opportunities. Especially the question of data access.

The current thesis investigates data-driven simulation decision-making with field-quality consumer data. This is accomplished by outlining the benefits and uses of combining machine learning and simulation in the literature and by locating barriers to the use of machine learning (ML) in the simulation subsystems at a case study organization. Additionally, an implementation is carried out to demonstrate how Scania departments can use this technology to analyze their current data and produce results that support the exploration of the simulation space and the identification of potential design issues so that preventative measures can be taken during concept development. The thesis' findings provide an overview of the literature on the relationship between machine learning and simulation technologies, as well as limitations of using machine learning in simulation systems at large scale manufacturing organizations. Support vector machines, logistic regression, and Random Forest classifiers are used to demonstrate one possible use of machine learning in simulation.

Datadriven kultur har flitigt nämnts i litteraturen som en tydlig framgångsfaktor för stora verksamheter för att skapa konkurrenskraft på marknaden.  Genom att verksamheter kan ta strategiska beslut baserat på stora mängder data förankrad i verkligheten undviks beslut som tas på magkänsla, således leder till optimering av verksamheter. Dock har små och medelstora företag (SMFs) halkat efter i utvecklingen då verksamheterna ofta saknar resurser och kompetens för att möjliggöra en datadriven kultur. Syftet med forskningen är att identifiera framgångsfaktorer speciellt inriktade på SMFs och skapa en sammanställning som SMF kan ta del av för att skapa en datadriven kultur. Den primära datainsamlingen genomfördes genom en kvalitativa ansats och fallstudie som forskningsmetod med semi-strukturerade intervjuer inriktade mot IT-branschen inom SMF som besatt på relevant kunskap inom ämnesområdet. Respondenternas svar har analyserats i jämförelse med tidigare litteratur för att generera framgångsfaktorer som möjliggör en datadriven kultur hos SMFs. Resultatet av forskningen har genererat en sammanställning på totalt fyra bekräftade huvudkategorier och sexton bekräftade underkategorier varav åtta berikande underkategorier är nya framgångsfaktorer som uppkommit från intervjuerna. De identifierade framgångsfaktorerna kan anammas av SMF för att möjliggöra den digitala transformationen mot en datadriven kultur. Resultatet av forskningen illustrerar att SMFs har stora möjligheter att öka sin konkurrenskraft, affärsvärde och produktivitet genom att tillämpa framgångsfaktorerna inom SMF och att en datadriven kultur inte är begränsade till stora verksamheter. / Data-driven culture has frequently been mentioned in the literature as a clear success factor for large enterprises (LEs) creating competitive advantages in the market. As enterprises can make strategic decisions based on large amounts of data anchored in reality, decisions are based on gut feeling, thus leading to optimization of enterprises. However, small and medium-sized enterprises (SMEs) have fallen behind in development as the enterprises often lack resources and knowledge to enable a data-driven culture. The purpose of the research is to identify success factors specifically focused on SMEs and create a compilation of which SMEs can adopt to create a data-driven culture. The primary data collection was conducted with a qualitative approach carrying out a case study with semi-structured interviews focused on the IT industry within SMEs that are obsessed with relevant knowledge in the subject area. The interviewees' responses have been analyzed in comparison with previous literature to generate success factors that enable a data-driven culture in SMEs. The results of the research have generated a compilation of a total of four confirmed main categories and sixteen confirmed subcategories, of which eight enriching subcategories are new success factors that have emerged from the interviews. The identified success factors can be adopted by SMEs to enable the digital transformation towards a data-driven culture. The results of the research illustrates that SMEs have great opportunities to increase in competitive advantages, business value and productivity by applying the success factors within SMEs and that the data-driven culture is not limited to LE.

Vehicle activated signs (VAS) display a warning message when drivers exceed a particular threshold. VAS are often installed on local roads to display a warning message depending on the speed of the approaching vehicles. VAS are usually powered by electricity; however, battery and solar powered VAS are also commonplace. This thesis investigated devel-opment of an automatic trigger speed of vehicle activated signs in order to influence driver behaviour, the effect of which has been measured in terms of reduced mean speed and low standard deviation. A comprehen-sive understanding of the effectiveness of the trigger speed of the VAS on driver behaviour was established by systematically collecting data. Specif-ically, data on time of day, speed, length and direction of the vehicle have been collected for the purpose, using Doppler radar installed at the road. A data driven calibration method for the radar used in the experiment has also been developed and evaluated. Results indicate that trigger speed of the VAS had variable effect on driv-ers’ speed at different sites and at different times of the day. It is evident that the optimal trigger speed should be set near the 85th percentile speed, to be able to lower the standard deviation. In the case of battery and solar powered VAS, trigger speeds between the 50th and 85th per-centile offered the best compromise between safety and power consump-tion. Results also indicate that different classes of vehicles report differ-ences in mean speed and standard deviation; on a highway, the mean speed of cars differs slightly from the mean speed of trucks, whereas a significant difference was observed between the classes of vehicles on lo-cal roads. A differential trigger speed was therefore investigated for the sake of completion. A data driven approach using Random forest was found to be appropriate in predicting trigger speeds respective to types of vehicles and traffic conditions. The fact that the predicted trigger speed was found to be consistently around the 85th percentile speed justifies the choice of the automatic model.