Digital Transformation (DX) has become a pervasive global phenomenon that is having a profound effect at all levels of an enterprise. The umbrella term DX is supported by nascent technologies such as blockchain, Internet of Things (IoT), and cloud-based computing and networking. The speed by which these new DX technologies have emerged has challenged current technical infrastructures, budgets, and skillsets as organizations attempt to incorporate and implement these technologies as part of pervasive digital operational initiatives. DX also includes the transformative effects (deployment and adoption) of these technologies, and these are referred to as “outcomes” in this thesis. Not surprisingly, recent studies indicate that at least seventy percent of DX projects either fail or underperform. As firms assess existing technology-to-business alignments in search of attributable causation, they discover that these alignments are often opaque regarding the capabilities required to obtain optimal digital transformation outcomes from the application of specific technologies. This is especially true as underlying technology infrastructures and architectures, which have had a traditionally functional role such as the data network, are increasingly relied upon to support the strategic outcome requirements of DX. This dissertation uses an inductive, multiple case study approach to explore these relationships and outcomes. It directly observes a set of large organizations across multiple verticals. These organizations have all completed pervasive digital transformation initiatives, more specifically, this study measured the resultant levels of digital transformation project outcomes achieved as a fraction of the initial digital transformation project deliverables. Furthermore, the study makes inferences regarding the relationship of the data network deployment posture to the observed digital project outcomes. To study this empirically, data network alignment is classified as either being functionally or strategically aligned. Analysis of the resultant data revealed four distinct themes: (a) organizations that procured data network equipment with pricing as the critical determinant experienced sub-par digital transformation outcomes; (b) organizations which considered key business drivers when procuring and architecting the data network achieved more successful digital transformation outcomes; (c) organizations which did not perform a network upgrade or perform a significant data network architectural change during the previous five years did not meet their own goals of digital transformation, and (d) some DX technology deployments achieved a high percentage of project deliverables without undertaking a data network upgrade. With respect to the fourth theme, however, the resulting low internal adoption of deployed DX technology among targeted operational units resulted in a ‘split-brain’ operation and thus an overall underperformance of the DX project. The final thesis chapter includes recommendations for future qualitative research on both digital transformation outcomes and the effect of managed network services on digital transformation outcomes as well as the need for quantitative research to establish deeper causation beyond the loose causation which is posited in this paper. / Business Administration/Management Information Systems

Last decade witnessed the explosive growth in mobile devices and their traffic demand, and hence the significant increase in the energy cost of the cellular service providers. One major component of energy expenditure comes from the operation of base stations. How to reduce energy cost of base stations while satisfying users’ soaring demands has become an imperative yet challenging problem. In this dissertation, we investigate the minimization of the long-term time-averaged expected energy cost while guaranteeing network strong stability. Specifically, considering flow routing, link scheduling, and energy constraints, we formulate a time-coupling stochastic Mixed-Integer Non-Linear Programming (MINLP) problem, which is prohibitively expensive to solve. We reformulate the problem by employing Lyapunov optimization theory and develop a decomposition based algorithm which ensures network strong stability. We obtain the bounds on the optimal result of the original problem and demonstrate the tightness of the bounds and the efficacy of the proposed scheme.

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第23550号 / 情博第780号 / 新制||情||133(附属図書館) / 京都大学大学院情報学研究科通信情報システム専攻 / (主査)教授 大木 英司, 教授 原田 博司, 教授 湊 真一 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

Computer networks are continuously evolving and growing in size and complexity. New technologies are being introduced which further increases the complexity. Net- work Service Orchestration is all about pushing configuration out into the network devices automatically without human intervention. There can be issues that causes the orchestration to fail. In many cases manual operations must be done to recover from the error which is very contradicting since the goal of orchestration is that it should be fully automated. There is some indication that the errors that are being solved manually could be de- tected and handled by a feedback mechanism. This thesis work aimed to build on current insight and if possible, verify that the feedback mechanism is a viable method. After consideration on different ways to solve the research question, the choice fell on creating a test environment where the approach was tested. The test environment was used to investigate if a network orchestration system could be integrated with a feedback mechanism. The result of this project presents a way to automatically de- tect a network failure and send feedback to a Network Service Orchestrator. The or- chestrator is then able to identify and correct the error. / Datornätverk utvecklas kontinuerligt och växer i storlek och komplexitet. Nyteknik införs som ytterligare ökar komplexiteten. Nätverksservice orkestrering handlar om att skicka ut konfiguration automatiskt till enheter i nätverket utan mänsklig in- blandning. Det kan finnas problem som gör att orkestreringen misslyckas. I många fall måste manuella åtgärder utföras för att lösa problemet, vilket är mycket motsä- gelsefullt, eftersom målet med orkestrering är att det ska vara fullt automatiserat. Det finns indikationer på att fel kan detekteras och hanteras av en återkopplings- mekanismen. Detta examensarbete syftar till att bygga på aktuell insikt, och om möj- ligt, verifiera att återkopplingsmekanismen är en möjlig metod. Efter överväganden på vilka olika sätt som projektmålet kunde uppnås föll valet på att skapa en testmiljö där ansatsen kunde testas. Testmiljön användes för att utreda om ett nätverksorkestreringssystem kan integreras med en återkopplings mekanism. Resultat av projektet presenterar ett sätt att automatiskt upptäcka ett nätverksfel och skicka återkoppling till ett nätverksorkestreringssystem. Nätverksorkestreraren kan sedan detektera och åtgärda felet.

Although generative adversarial networks (GANs) have achieved state-of-the-art results in generating realistic look- ing images, they are often parameterized by neural net- works with relatively few learnable weights compared to those that are used for discriminative tasks. We argue that this is suboptimal in a generative setting where data is of- ten entangled in high dimensional space and models are ex- pected to benefit from high expressive power. Additionally, in a generative setting, a model often needs to extrapo- late missing information from low dimensional latent space when generating data samples while in a typical discrimina- tive task, the model only needs to extract lower dimensional features from high dimensional space. We evaluate different architectures for GANs with varying model capacities using shortcut connections in order to study the impacts of the capacity on training stability and sample quality. We show that while training tends to oscillate and not benefit from additional capacity of naively stacked layers, GANs are ca- pable of generating samples with higher quality, specifically for images, samples of higher visual fidelity given proper regularization and careful balancing. / Trots att Generative Adversarial Networks (GAN) har lyckats generera realistiska bilder består de än idag av neurala nätverk som är parametriserade med relativt få tränbara vikter jämfört med neurala nätverk som används för klassificering. Vi tror att en sådan modell är suboptimal vad gäller generering av högdimensionell och komplicerad data och anser att modeller med högre kapaciteter bör ge bättre estimeringar. Dessutom, i en generativ uppgift så förväntas en modell kunna extrapolera information från lägre till högre dimensioner medan i en klassificeringsuppgift så behöver modellen endast att extrahera lågdimensionell information från högdimensionell data. Vi evaluerar ett flertal GAN med varierande kapaciteter genom att använda shortcut connections för att studera hur kapaciteten påverkar träningsstabiliteten, samt kvaliteten av de genererade datapunkterna. Resultaten visar att träningen blir mindre stabil för modeller som fått högre kapaciteter genom naivt tillsatta lager men visar samtidigt att datapunkternas kvaliteter kan öka, specifikt för bilder, bilder med hög visuell fidelitet. Detta åstadkoms med hjälp utav regularisering och noggrann balansering.

The Internet and intranets are viewed as capable of supplying "Anything, Anywhere, Anytime" and e-commerce, e-government, e-community, and military C4I are now deploying many and varied applications to serve their needs. Network management is currently centralized in operations centers. To assure customer satisfaction with the network performance they typically plan, configure and monitor the network devices to insure an excess of bandwidth, that is overprovision. If this proves uneconomical or if complex and poorly understood interactions of equipment, protocols and application traffic degrade performance creating customer dissatisfaction, another more application-centric, way of managing the network will be needed. This research investigates a new qualitative class of network performance measures derived from the current quantitative metrics known as quality of service (QOS) parameters. The proposed class of qualitative indicators focuses on utilizing current network performance measures (QOS values) to derive abstract quality of experience (QOE) indicators by application class. These measures may provide a more user or application-centric means of assessing network performance even when some individual QOS parameters approach or exceed specified levels. The mathematics of functional analysis suggests treating QOS performance values as a vector, and, by mapping the degradation of the application performance to a characteristic lp-norm curve, a qualitative QOE value (good/poor) can be calculated for each application class. A similar procedure could calculate a QOE node value (satisfactory/unsatisfactory) to represent the service level of the switch or router for the current mix of application traffic. To demonstrate the utility of this approach a discrete event simulation (DES) test-bed, in the OPNET telecommunications simulation environment, was created modeling the topology and traffic of three semi-autonomous networks connected by a backbone. Scenarios, designed to degrade performance by under-provisioning links or nodes, are run to evaluate QOE for an access network. The application classes and traffic load are held constant. Future research would include refinement of the mathematics, many additional simulations and scenarios varying other independent variables. Finally collaboration with researchers in areas as diverse as human computer interaction (HCI), software engineering, teletraffic engineering, and network management will enhance the concepts modeled.

A reconfiguration algorithm for electrical distribution system to reduce system losses is presented. The algorithm determines the switching patterns as a function of time. Either seasonal or daily time studies may be performed. Both manual and automatic switches are used to reconfigure the system for seasonal studies, whereas only automatic switches are considered for daily studies. An algorithm for load estimation is developed. The load estimation algorithm provides load information for each time point to be analyzed. The load estimation algorithm can incorporate any or all of the following: spot loads, circuit measurements, and customer time-varying diversified load characteristics. Voltage dependency of loads is considered at the circuit level. It is shown that switching at the system peak can reduce losses but may cause a marginal increase in system peak. Voltage and current constraints are incorporated in the reconfiguration algorithm. Data base tables and data structures used in the algorithm are described. Example problems are provided to illustrate results. / Ph. D.

Traditional call button networks that control elevator systems utilize a wired connection for communication. The communication cables are run through the elevator shaft from one call button to another and finally to the controller on the roof. Installing this wired link is highly time consuming. In this thesis, we propose the design for a wireless call button network. Two important features of this wireless network design are low cost and low power consumption. Controller Area Network (CAN) is a widely used protocol for wired networks and has been proposed for use in next generation elevator control systems. A modified CAN for wireless (MCANW) protocol has been developed for the wireless call button network. The wireless link will be implemented via the use of data radios. A modified form of traditional Binary Phase Shift Keying (BPSK) modulation scheme for the radios is proposed. The proposed modulation scheme, like differential BPSK, can be detected non-coherently but it offers better performance than differential BPSK. Its implementation includes an innovative tracking algorithm to maintain synchronization at the receiver. / Master of Science

The Winner-Take-All (WTA) is an essential neural network operation for locating the most active neuron. Such a procedure has been extensively used in larger application areas. The Winner-Take-All circuit selects the maximum of the inputs inhibiting all other nodes. The efficiency of the analog circuits may well be considerably higher than the digital circuits. Also, analog circuits’ design footprint and processing time can be significantly small. A current-based Winner-Take-All circuit for analog neural networks is presented in this research. A compare and pass (CAP) mechanism has been used, where each input pair is compared, and the winner is selected and passed to another level. The inputs are compared by a sense amplifier which generates high and low voltage signals at the output node. The voltage signal of the sense amplifier is used to select the winner and passed to another level using logic gates. Also, each winner follows a sequence of digital bits to be selected. The findings of the SPICE simulation are also presented. The simulation results on the MNIST, Fashion-MNIST, and CIFAR10 datasets for the memristive deep neural network model show the significantly accurate result of the winner class with an average difference of input and selected winner output current of 0.00795uA, 0.01076uA and 0.02364uA respectively. The experimental result with transient noise analysis is also presented.

High-speed transmission over contemporary communication networks has drawn many research efforts. Traffic scheduling schemes which play a critical role in managing network transmission have been pervasively studied and widely implemented in various practical communication networks. In a sophisticated communication system, a variety of applications co-exist and require differentiated Quality-of-Service (QoS). Innovative scheduling schemes and hybrid scheduling disciplines which integrate multiple traditional scheduling mechanisms have emerged for QoS differentiation. This study aims to develop novel analytical models for commonly interested scheduling schemes in communication systems under more realistic network traffic and use the models to investigate the issues of design and development of traffic scheduling schemes. In the open literature, it is commonly recognized that network traffic exhibits self-similar nature, which has serious impact on the performance of communication networks and protocols. To have a deep study of self-similar traffic, the real-world traffic datasets are measured and evaluated in this study. The results reveal that selfsimilar traffic is a ubiquitous phenomenon in high-speed communication networks and highlight the importance of the developed analytical models under self-similar traffic. The original analytical models are then developed for the centralized scheduling schemes including the Deficit Round Robin, the hybrid PQGPS which integrates the traditional Priority Queueing (PQ) and Generalized Processor Sharing (GPS) schemes, and the Automatic Repeat reQuest (ARQ) forward error control discipline in the presence of self-similar traffic. Most recently, research on the innovative Cognitive Radio (CR) techniques in wireless networks is popular. However, most of the existing analytical models still employ the traditional Poisson traffic to examine the performance of CR involved systems. In addition, few studies have been reported for estimating the residual service left by primary users. Instead, extensive existing studies use an ON/OFF source to model the residual service regardless of the primary traffic. In this thesis, a PQ theory is adopted to investigate and model the possible service left by selfsimilar primary traffic and derive the queue length distribution of individual secondary users under the distributed spectrum random access protocol.