No description available.

This thesis studies the generalization ability of machine learning algorithms in a statistical setting. It focuses on the data-dependent analysis of the generalization performance of learning algorithms in order to make full use of the potential of the actual training sample from which these algorithms learn.¶ First, we propose an extension of the standard framework for the derivation of generalization bounds for algorithms taking their hypotheses from random classes of functions. This approach is motivated by the fact that the function produced by a learning algorithm based on a random sample of data depends on this sample and is therefore a random function. Such an approach avoids the detour of the worst-case uniform bounds as done in the standard approach. We show that the mechanism which allows one to obtain generalization bounds for random classes in our framework is based on a “small complexity” of certain random coordinate projections. We demonstrate how this notion of complexity relates to learnability and how one can explore geometric properties of these projections in order to derive estimates of rates of convergence and good confidence interval estimates for the expected risk. We then demonstrate the generality of our new approach by presenting a range of examples, among them the algorithm-dependent compression schemes and the data-dependent luckiness frameworks, which fall into our random subclass framework.¶ Second, we study in more detail generalization bounds for a specific algorithm which is of central importance in learning theory, namely the Empirical Risk Minimization algorithm (ERM). Recent results show that one can significantly improve the high-probability estimates for the convergence rates for empirical minimizers by a direct analysis of the ERM algorithm. These results are based on a new localized notion of complexity of subsets of hypothesis functions with identical expected errors and are therefore dependent on the underlying unknown distribution. We investigate the extent to which one can estimate these high-probability convergence rates in a data-dependent manner. We provide an algorithm which computes a data-dependent upper bound for the expected error of empirical minimizers in terms of the “complexity” of data-dependent local subsets. These subsets are sets of functions of empirical errors of a given range and can be determined based solely on empirical data. We then show that recent direct estimates, which are essentially sharp estimates on the high-probability convergence rate for the ERM algorithm, can not be recovered universally from empirical data.

The goal of this research was to improve ocean colour chlorophyll a (Chla) retrievals in the coastal Case 2 waters of the Salish Sea by characterizing the main drivers of optical variability and using this information to parameterize empirical algorithms based on an optical classification. This was addressed with three specific objectives: (1) build a comprehensive spatio-temporal data set of in situ optical and biogeochemical parameters, (2) apply a hierarchical clustering analysis to classify above-water remote sensing reflectance (Rrs) and associated bio-optical regimes, (3) optimize and validate class-specific empirical algorithms for improved Chla retrievals. Biogeochemical and optical measurements, acquired at 145 sites, showed considerable variation; Chla (mean=1.64, range: 0.10 – 7.20 µg l-1), total suspended matter (TSM) (3.09, 0.82 – 20.69 mg l-1), and absorption by chromophoric dissolved organic matter (a_cdom (443)) (0.525, 0.007 – 3.072 m-1), thus representing the spatial and temporal variability of the Salish Sea. A comparable range was found in the measured optical properties; particulate scattering (b_p (650)) (1.316, 0.250 – 7.450 m-1), particulate backscattering (b_bp (650)) (0.022, 0.005 – 0.097 m-1), total beam attenuation coefficient (c_t (650)) (1.675, 0.371 – 9.537 m-1), and particulate absorption coefficient (a_p (650)) (0.345, 0.048 – 2.020 m-1). Empirical orthogonal function (EOF) analysis revealed 95% of the Rrs variance was highly correlated to b_p (r = 0.90), b_bp (r = 0.82), and TSM concentration (r = 0.80), suggesting a strong influence from riverine systems in this region. Hierarchical clustering on the normalized Rrs revealed four spectral classes. Class 1 is defined by high overall Rrs magnitudes in the red, indicating more turbid waters, Class 2 showed high Rrs values in the red and well defined fluorescence and absorption features, indicated by a high Chla and TSM presence, Class 3 showed low TSM influence and more defined Chla signatures, and Class 4 is characterized by overall low Rrs values, suggesting more optically clear oceanic waters. Spectral similarities justified a simplification of this classification into two dominant water classes – (1) estuarine class (Classes 1 and 2) and (2) oceanic class (Classes 3 and 4) – representing the dominant influences seen here. In situ Chla and above-water remote sensing reflectance measurements, used to validate and parameterize the OC3M/OC3S3, two-band ratio, FLH and, modified FLH (ModFLH) empirical algorithms, showed a systematic overestimation of low Chla concentrations and underestimation of higher Chla values for all four algorithms when tuned to regional data. FLH and ModFLH algorithms performed best for these data (R2 ~ 0.40; RMSE ~ 0.32). Algorithm accuracy was significantly improved for the class-specific parametrizations with the two-band ratio showing a strong correlation to the Chla concentrations in the estuarine class (R2 ~ 0.71; RMSE ~ 0.33) and the ModFLH algorithm in the oceanic class (R2 ~ 0.70; RMSE ~ 0.26). These results demonstrated the benefit of applying an optical classification as a necessary first step into improving Chla retrievals from remotely sensed data in the contrasted coastal waters of the Salish Sea. With accurate Chla information, the health of the Salish Sea can be viably monitored at spatial and temporal scales suitable for ecosystem management. / Graduate / 0416 / stephen.uvic@gmail.com

This study investigates the experience of exchange studies from a phenomenological perspective. Phenomenology studies the sentence structure of the phenomenon and places the lifeworld as a central concept. Husserl (2002/1954) described the lifeworld as a coherent universe of objects where we, all of us living with another in this world, may be conscious of the world as a universal horizon. In this study the lifeworld of exchange students is investigated with emphasis on their experience of the exchange. The aim of the study is to highlight the sentence structure in the experience of living as an exchange student which involves a meeting with a different culture. Seven interviews were conducted with exchange students that participated in at least one semester of higher education abroad and the resulting data was analyzed using the EPP method. This method describes the general meaning structure of the phenomenon and identifies several distinguishing features. For this study ten features were identified; an expectation and openness relative an uncertain future, a deliberate break, a part of something bigger, a deliberate approach to cultural codes, an encounter with different values, new interactions and relations, language difficulties, a new environment, a creation of identity and a personal development and responsibilities. The results indicate that the experience of exchange studies may not always be unproblematic since it’s often difficult to gain access to the culture and society of the host country. However, the experience also seems to be rewarding and is characterized by a personal development and a change in the perception of both oneself and people from other cultures. The results also show that the exchange student may gain a feeling of living in a global society where all possibilities are present. / Denna studie utgår ifrån en fenomenologisk ansats. Fenomenologins uppgift är att beskriva ett fenomens meningsstruktur där ett centralt begrepp är livsvärlden. Enligt Husserl (2002/1954) belyser livsvärlden den verklighet som man ständigt befinner sig i tillsammans med andra människor. Grundvalen för denna studie är att studera utbytesstudenters livsvärld där upplevelsen av utbytesstudier är i fokus. Syftet med studien är att beskriva meningsstrukturer i upplevelsen av att leva som utbytesstudent och som inbegriper ett möte med en annan kultur. Sju intervjuer har genomförts med utbytesstudenter som alla studerat minst en termin inom högre utbildning vid ett värduniversitet i ett annat land. De aktuella data som behandlats har analyserats med hjälp av studiens valda metod, Empirical phenomenological psychologial (EPP-metoden). Resultatet beskriver det aktuella fenomenets meningsstruktur som en helhet och visar vilka de generella kännetecknen är. De generella kännetecknen som framkommit i studien är; en förväntan och en öppenhet inför en obestämd framtid, ett medvetet uppbrott, en del i att uppnå något större, ett medvetet förhållningssätt till kulturella koder, ett möte med andra värderingar, nya interaktioner och relationer, svårigheter med språket, ändrade livsvillkor, ett identitetsskapande, en personlig utveckling och ett ansvarstagande. Resultatet indikerar att upplevelsen av utbytesstudier och att möta en annan kultur inte är helt oproblematisk då det ibland är svårt att få tillträde till värdlandets kultur och samhälle. Dock verkar upplevelsen även vara en värdefull tid som präglas av en utveckling och ett ifrågasättande av tidigare föreställningar om en själv och människor från andra kulturer. Resultatet påvisar också att utbytesstudenten kan ha en känsla av att leva i en global värld där alla möjligheter finns.

Context: Global software development is the emerging trend in today’s world as it provides the software companies with certain advantages like access to skill and cheap labor, low development cost etc. There are also many challenges and risks involved in globally distributed software project as compare to co-located projects like communication, coordination and control etc. due to geographical separation. There is also a recent interest in applying agile methods in offshore projects in order to minimize certain offshore challenges. Agile methods in collocated environment demands high communication, coordination and collaboration between the team members. However implementing agile methods in distributed environment is beneficial and challenging too, because of inherited GSD challenges. Therefore, it is necessary to carefully understand the benefits and challenges of applying agile methods in distributed project before actually initiating the development. Objectives: This research study aims to look at current empirical evidence regarding benefits and challenges of applying agile methods in offshore projects, in order to understand it properly. Furthermore, we have examined that which of the benefits and challenges reported in literature can be seen in practice. This helped us in finding out the similarities and differences between the benefits and challenges that are reported in literature and in practice. Methods: To fulfill our aims and objectives we have first conducted the detailed systematic review analysis of the empirical studies from year 2000-2010. The systematic review approach helped us to collect and summarize the empirical data available regarding the benefits and challenges of applying agile methods in offshore projects. The research studies were identified from the most authentic databases that are scientifically and technically peer reviewed such as Inspec and Compendex, IEEE Xplore, ACM digital library, Springer Link, Science Direct, Google Scholar, Scopus, ISI Web of Knowledge and Willey Inter Science Journal Finder. Furthermore, industrial survey was conducted in order to identify whether the benefits and challenges enlisted in literature can be confirmed by practitioners. This helped us in determining the similarities and differences between benefits and challenges that are reported in empirical literature and industrial survey. Results: From systematic review results, we have found the list of benefits and challenges of most applied agile practices such as daily scrum meetings, sprint planning meetings, sprint review meetings, retrospective meetings, pair programming, short release and on-site customer. The most common benefits related to above identified agile practices are early problem identification, project visibility, increase communication, coordination and collaboration, trust, clarification of work and requirements, early client feedback and customer satisfaction. Similarly the most common challenges associated with above identified agile practices are lack of linguistic skills, culture differences and temporal differences. From survey result, it was found that the majority of benefits and challenges identified through literature review are prevalent in software industries to some extent. We were able to identify the benefits and challenges of agile practices such as daily scrum, sprint planning meetings, sprint review, retrospective meetings and short release. The most common benefits of above identified agile practices highlighted by survey participants are project visibility, early problem identification, increase coordination, opportunity for knowledge sharing and transferring, immediate feedback from client, and opportunity for understanding the task properly. Similarly the most common challenges identified by survey participants are language problems, temporal difference and difficult to explain and understand problems or issues over communication technologies. However, there are certain challenges like trust and culture issues, which have received most attention in the empirical literature, but are not identified by majority of industry respondents. Conclusions: From the information attained so far regarding the benefits and challenges of agile methods in offshore projects, we observed that there is an emerging trend of using agile methods in offshore projects and practitioners actually found it beneficial. The benefits of agile practices does not achieve straightaway there are certain challenging factors of GSD too, which can restrict the implementation or make the methodology not useful such as communication, coordination and collaboration. These challenges are usually occurred due to geographical separation, lack of trust and linguistic skills, culture difference etc. The presented work add contribution in a sense as to the best of our knowledge no systematic review effort has been done in this area with specifically focuses on highlighting the benefits and challenges. From survey results we found that majority of benefits and challenges highlighted by survey participants are to some extent same with what we have already studied in literature. However there are certain important challenges which have received most attention in empirical literature but were less identified by the survey participants like trust issue and culture difference. The benefits and challenges identified through industrial survey do not add much information into our attention. This does not mean that all possible benefits and challenges of applying agile methods in offshore project have been explored. The responses which were received from industry practitioners were quite few to make any kind of arguments. There is a probability that we missed the important benefits and challenges related to particular agile practices during our industrial survey. Therefore we conclude that a much thorough and inclusive answers should be required to identify the benefits and challenges of applying agile methods in offshore projects.

Context. Software Product Management (SPM) is an activity done by software product managers to develop products from the very first idea till the product is release to the market and providing service to customers. SPM helps software product managers to successfully maintain their product throughout product lifecycle till it is disposed of from the market. In this thesis we have studied about SPM both from the state of knowledge and state of practice to understand the different activities being followed by software product managers and concepts in those activities. Objectives. The study identifies the practices followed in SPM, both from the literature and industrial practitioners. The study also unravels the order in which they implement SPM activities. We have focused on what activities need to be focused by junior software product manager when implementing SPM and based on the inputs from software practitioners a preliminary framework was designed. Methods. Literature review was conducted to identify what are the different practices in SPM. To cover the literature related to SPM different international databases like Compendex, Inspec, IEEE Xplore, ACM Digital Library, Wiley Inter Science, Scopus, Science Direct, and Business Source Premier are selected. Articles from databases are selected between the years 1995-2011. Text books related to SPM, conference proceedings and grey literature are also used. Interviews and Survey are conducted with practitioners to identify how SPM is being implemented in organizations. We have also focused on what activities need to be implemented in SPM by junior software product managers after they take up their job. Results. Totally 133 different practices are identified related to SPM and are grouped in to their respective activities. Based on the findings a preliminary framework was designed for junior software product managers where they need to implement 22 concepts and these concepts are grouped into their respective activities which are Requirements Management, Release Planning, Product Planning, Product Roadmapping and Marketing. Conclusions. We conclude that the SPM frameworks identified through literature relate to each other and use similar activities. We also conclude that the frameworks identified from literature cover almost all of the activities implemented by software product managers during SPM when empirical research was conducted. The frameworks identified from literature don’t give a clear picture about what activities need to be focused by junior software product managers when implementing SPM. To overcome this, a preliminary framework was designed for junior software product managers based on the empirical evidence identified from interviews and survey. Moreover we conclude that a curriculum needs to be designed for SPM which clearly teaches junior software product managers regarding the activities and concepts involved in SPM and focusing more on the activities which need to be implemented by junior software product managers.

Maritime transportation emits about 3% of global greenhouse gas, International Maritime Organization (IMO) aims to reduce shipping’s emissions by 50% with respect to 2008 levels. Proton exchange membrane fuel cells (PEMFCs) are considered among the most promising clean technologies for decarbonizing the maritime sector. One of the challenges for commercial application of PEMFCs is their limited durability. The purpose of this thesis was to assess the most significant degradation mechanisms and operating conditions of the PEMFC in marine applications, including membrane and catalyst layer degradation during idle, start-stop cycles, and dynamic load cycles, and to build a model to forecast the lifetime.A semi-empirical approach was developed to evaluate the PEMFC lifetime through a 2D COMSOL model. The model takes into account the empirical relationships for membrane conductivity loss and electrochemical surface area (ECSA) decay as functions of cycling numbers, aging process, and idling time. The 2D model has been validated with the experimental data in the literature and are also compared with a previous 1D model. The polarization curves show the voltage output against current density, lifetime is evaluated using a 10% voltage reduction criterion at the current density 0.6 A/cm2.An improved ECSA degradation model with variable load levels increases the lifetime of the ferry in Case 5 from 5500 hours to 7500 hours. Load cycling and idling cause the most severe degradation, but the impact can be reduced by a hybrid system with battery supplement and onshore charging. The lifetime of the ferry in Case 5 has been significantly further improved from 7500 hours to 22500 hours, which is comparable to the 20000-hour lifetime of commercial products for marine applications. Furthermore, membrane thickness effect analysis showed that fuel cells with thinner membranes (such as NR211) have better performance before degradation due to higher proton conductivity, but degrade faster during load cycling due to hydrogen crossover. The results of this research can be extended to help optimize fuel cell, stack and power system designs to avoid worst-case operating conditions and thereby limit fuel cell degradation. / Sjötransporter släpper ut cirka 3% av de globala växthusgaserna, International Maritime Organization (IMO) har som mål att minska sjöfartens utsläpp med 50 % jämfört med 2008 års nivåer. PEM-bränsleceller anses vara bland de mest lovande rena teknikerna för att minska koldioxidutsläppen i den maritima sektorn. En av utmaningarna för kommersiell användning av PEM-bränsleceller är deras begränsade hållbarhet. Syftet med denna avhandling var att bedöma de viktigaste nedbrytningsmekanismerna och driftsförhållandena för PEM-bränsleceller i marina applikationer, inklusive nedbrytning av membran och katalysatorskikt under tomgång, start-stopp-cykler och dynamiska belastningscykler, och att bygga en modell för att förutsäga livslängd.En semi-empirisk metod utvecklades för att utvärdera PEMFC:s livslängd genom en 2D COMSOL-modell. Modellen tar hänsyn till de empiriska sambanden för membrankonduktivitetsförlust och den elektrokemisk ytareans (ECSA) sönderfall som funktioner av cyklingstal, åldrandeprocess och tomgångstid. 2D-modellen har validerats med experimentella data i litteraturen och jämförs även med en tidigare 1D-modell. Polarisationskurvorna visar utspänningen mot strömtätheten, livslängden utvärderas med ett 10 % spänningsreduktionskriterium vid strömtätheten 0.6 A/cm2.En förbättrad modell för nedbrytning av elektrokemisk yta med varierande lastnivåer ökar färjans livslängd i fall 5 från 5500 timmar till 7500 timmar. Lastcykling och tomgång orsakar den allvarligaste försämringen, men påverkan kan minskas genom ett hybridsystem med batteritillägg och landladdning. Färjans livslängd i fall 5 har förbättrats avsevärt ytterligare från 7500 timmar till 22500 timmar, vilket är jämförbart med 20000 timmars livslängd för kommersiella produkter för marina applikationer. Vidare visade membrantjocklekseffektanalys att bränsleceller med tunnare membran (som NR211) har bättre prestanda före nedbrytning på grund av högre protonledningsförmåga, men bryts ned snabbare under belastningscykler på grund av väteövergång. Resultaten av denna forskning kan utökas för att hjälpa till att optimera designen av bränsleceller, stack och kraftsystem för att undvika värsta driftsförhållanden och därigenom begränsa nedbrytningen av bränsleceller.

<p>    </p> <p>Social Media Platforms (SMPs) are used by almost 60% of the global population. Along with the ubiquity of SMPs, there are increasing Trust & Safety (T&S) risks that expose users to spam, harassment, abuse, and other harmful content online. <em>T&S Engineering </em>is an emerging area of software engineering striving to mitigate these risks. This study provides the first step in understanding this form of software engineering.</p> <p>This study examines how T&S Engineering is practiced by SMP engineers. I studied two open-source (OSS) SMPs, Mastodon and Diaspora, which comprise 89% of the 9.6 million OSS SMP accounts. I focused on the T&S design process by analyzing T&S discussions within 60 GitHub issues. I applied a T&S discussion model to taxonomize the T&S risks, T&S engineering patterns, and resolution rationales. I found that T&S issues persist throughout a platform’s lifetime, they are difficult to resolve, and engineers favor reactive treatments. To integrate findings, I mapped T&S engineering patterns onto a gen- eral model of SMPs. My findings give T&S engineers a systematic understanding of their T&S risk treatment options. I conclude with future directions to study and improve T&S Engineering, spanning software design, decision-making, and validation. </p>

This report evaluates the possibility of using sequential learning in a material development setting to help predict material properties and speed up the development of new materials. To do this a Random forest model was built incorporating carefully calibrated prediction uncertainty estimates. The idea behind the model is to use the few data points available in this field and leverage that data to build a better representation of the input-output space as each experiment is performed. Having both predictions and uncertainties to evaluate, several different strategies were developed to investigate performance. Promising results regarding feasibility and potential cost-cutting were found using these strategies. It was found that within a specific performance region of the output space, the mean difference in alloying component price between the cheapest and most expensive material could be as high as 100 %. Also, the model performed fast extrapolation to previously unknown output regions, meaning new, differently performing materials could be found even with very poor initial data. / I denna rapport utvärderas möjligheten att använda sekventiell maskininlärning inom materialutveckling för att kunna prediktera materials egenskaper och därigenom förkorta materialutvecklingsprocessen. För att göra detta byggdes en Random forest regressionsmodell som även innehöll en uppskattning av prediktionsosäkerheten. Tanken bakom modellen är att använda de relativt få datapunkter som generellt brukar vara tillgängliga inom materialvetenskap, och med hjälp av dessa bygga en bättre representation av input-output-rummet genom varje experiment som genomförs. Med både förutsägelser och osäkerheter att utvärdera utvecklades flera olika strategier för att undersöka prestanda för de olika kandidatmaterialen. Genom att använda dessa strategier kunde lovande resultat vad gäller genomförbarhet och potentiell kostnadsbesparing hittas. Det visade sig att, för specifika prestandakrav, den genomsnittliga skillnaden i pris mellan den billigaste och den dyraste materialkemin kan vara så hög som 100 %. Vad gäller övriga resultat klarade modellen av att snabbt extrapolera initial data till tidigare okända regioner av output-rummet. Detta innebär att nya material med ny typ av prestanda kunde hittas även med mycket missanpassad initial träningsdata.

This report evaluates the possibility of using sequential learning in a material development setting to help predict material properties and speed up the development of new materials. To do this a Random forest model was built incorporating carefully calibrated prediction uncertainty estimates. The idea behind the model is to use the few data points available in this field and leverage that data to build a better representation of the input-output space as each experiment is performed. Having both predictions and uncertainties to evaluate, several different strategies were developed to investigate performance. Promising results regarding feasibility and potential cost-cutting were found using these strategies. It was found that within a specific performance region of the output space, the mean difference in alloying component price between the cheapest and most expensive material could be as high as 100 %. Also, the model performed fast extrapolation to previously unknown output regions, meaning new, differently performing materials could be found even with very poor initial data. / I denna rapport utvärderas möjligheten att använda sekventiell maskininlärning inom materialutveckling för att kunna prediktera materials egenskaper och därigenom förkorta materialutvecklingsprocessen. För att göra detta byggdes en Random forest regressionsmodell som även innehöll en uppskattning av prediktionsosäkerheten. Tanken bakom modellen är att använda de relativt få datapunkter som generellt brukar vara tillgängliga inom materialvetenskap, och med hjälp av dessa bygga en bättre representation av input-output-rummet genom varje experiment som genomförs. Med både förutsägelser och osäkerheter att utvärdera utvecklades flera olika strategier för att undersöka prestanda för de olika kandidatmaterialen. Genom att använda dessa strategier kunde lovande resultat vad gäller genomförbarhet och potentiell kostnadsbesparing hittas. Det visade sig att, för specifika prestandakrav, den genomsnittliga skillnaden i pris mellan den billigaste och den dyraste materialkemin kan vara så hög som 100 %. Vad gäller övriga resultat klarade modellen av att snabbt extrapolera initial data till tidigare okända regioner av output-rummet. Detta innebär att nya material med ny typ av prestanda kunde hittas även med mycket missanpassad initial träningsdata.