This thesis is devoted to the development of extensions of dynamic programming to the study of decision trees. The considered extensions allow us to make multi-stage optimization of decision trees relative to a sequence of cost functions, to count the number of optimal trees, and to study relationships: cost vs cost and cost vs uncertainty for decision trees by construction of the set of Pareto-optimal points for the corresponding bi-criteria optimization problem. The applications include study of totally optimal (simultaneously optimal relative to a number of cost functions) decision trees for Boolean functions, improvement of bounds on complexity of decision trees for diagnosis of circuits, study of time and memory trade-off for corner point detection, study of decision rules derived from decision trees, creation of new procedure (multi-pruning) for construction of classifiers, and comparison of heuristics for decision tree construction. Part of these extensions (multi-stage optimization) was generalized to well-known combinatorial optimization problems: matrix chain multiplication, binary search trees, global sequence alignment, and optimal paths in directed graphs.

V současné době se zvyšuje potřeba rozvoje agilních výrobních postupů, které lze snadno přizpůsobit neustálému zavádění nových produktů na trh. Jednobodové inkrementální tváření je nový, inovativní a proveditelný tvářecí proces s jednoduchým uspořádáním. Proces se provádí při pokojové teplotě (tváření za studena) a vyžaduje CNC stroj, nástroj s kulovou hlavou a jednoduché příslušenství pro uchycení obrobku plechu. V samotném procesu jde o přírůstkové formování, řízené CNC programem. Plastická deformace je lokalizována pod formovacím nástrojem takže plech je tvářen souhrnem pohybů lokální plastické zóny. Tento proces je zdlouhavý a proto se hodí pouze pro prototypovou výrobu nebo pro malé výrobní dávky. Na druhé straně umožňuje vyšší tvářitelnost ve srovnání s konvenčními procesy tváření, umožňuje použití levných nástrojů a také je charakterizován krátkou dobou od návrhu po výrobu produktu. Tato práce je výsledkem mezinárodní spolupráce Danmarks Tekniske Universitet v Lyngby a Instituto Superior Técnico v Lisabonu. Práce začíná krátkým hodnocením dílčích tvářecích procesů, pokračuje představováním jednobodového inkrementálního tváření a identifikací jeho praktických aplikací. Teoretická část obsahuje přehled nového rámce pro jednobodové inkrementální tváření, který je vytvořen na základě analýzy styku třecích sil. Praktická část projektu poskytuje úplný popis experimentálních technik použitých pro charakterizaci materiálů a stanovení limitů tvářitelnosti, dále se analyzuje vliv různých vstupních parametrů procesu (poloměru nástroje, tepelné zpracování materiálu obrobku, druh maziva,...). Tato část také obsahuje přehled experimentálního uspořádaní procesu jednobodového inkrementálního tváření i krátký popis CAD / CAM vývoje tří testovacích modelů. Poté jsou popsány v samostatné kapitole výsledky pozorování a analýzy hlavních parametrů procesů, které ovlivňují tvařitelnostní limity v jednobodovém inkrementálním tváření v souvislosti s aplikovaným teoretickým rámcem. Výsledky experimentů z časti objasňují probíhající mezinárodní diskusi kolem tvářitelnosti mechanismu jednobodového inkrementálního tváření vzhledem k tradičním metodám tváření. Jako logické pokračování prováděných experimentů, byla práce rozšířena na více-stupňové jednobodové inkrementální tváření, které umožňuje tváření součástek (kalíšku) se svislými stěnami ve více krocích. Za účelem objasnění procesů spojených s tímhle procesem byly navrženy a ve čtyřech krocích vyrobeny dva experimentální modely. Hlavním přínosem této práce k více-stupňovému jednobodovému inkrementálnímu tváření byla úspěšná výroba součásti s nekruhovým průřezem a kolmými stěnami. S cílem aplikovat celkové znalosti získáných v předchozích částí práce byla vyrobena prototypová součást. Popis designu a vývoje prototypu je součástí práce. V neposlední řadě jsou celkové závěry uvedené v poslední kapitole. Předpokládá se, že tato práce přizpívá k lepšímu pochopení mechanismu jednobodového inkrementálního tváření.

京都大学 / 0048 / 新制・課程博士 / 博士(エネルギー科学) / 甲第17915号 / エネ博第287号 / 新制||エネ||60(附属図書館) / 30735 / 京都大学大学院エネルギー科学研究科エネルギー変換科学専攻 / (主査)教授 石山 拓二, 教授 星出 敏彦, 准教授 川那辺 洋 / 学位規則第4条第1項該当 / Doctor of Energy Science / Kyoto University / DGAM

In recent years, edge computing has become an increasingly popular computing paradigm to enable real-time data processing and mobile intelligence. Edge computing allows computing at the edge of the network, where data is generated and distributed at the nearby edge servers to reduce the data access latency and improve data processing efficiency. In addition, with the advance of Artificial Intelligence of Things (AIoT), not only millions of data are generated from daily smart devices, such as smart light bulbs, smart cameras, and various sensors, but also a large number of parameters of complex machine learning models have to be trained and exchanged by these AIoT devices. Classical cloud-based platforms have difficulty communicating and processing these data/models effectively with sufficient privacy and security protection. Due to the heterogeneity of edge elements including edge servers, mobile users, data resources, and computing tasks, the key challenge is how to effectively manage resources (e.g. data, services) and schedule tasks (e.g. ML/FL tasks) in the edge clouds to meet the QoS of mobile users or maximize the platform's utility. To that end, this dissertation studies joint resource management and task scheduling for mobile edge computing. The key contributions of the dissertation are two-fold. Firstly, we study the data placement problem in edge computing and propose a popularity-based method as well as several load-balancing strategies to effectively place data in the edge network. We further investigate a joint resource placement and task dispatching problem and formulate it as an optimization problem. We propose a two-stage optimization method and a reinforcement learning (RL) method to maximize the total utilities of all tasks. Secondly, we focus on a specific computing task, i.e., federated learning (FL), and study the joint participant selection and learning scheduling problem for multi-model federated edge learning. We formulate a joint optimization problem and propose several multi-stage optimization algorithms to solve the problem. To further improve the FL performance, we leverage the power of the quantum computing (QC) technique and propose a hybrid quantum-classical Benders' decomposition (HQCBD) algorithm as well as a multiple-cuts version to accelerate the convergence speed of the HQCBD algorithm. We show that the proposed algorithms can achieve the consistent optimal value compared with the classical Benders' decomposition running in the classical CPU computer, but with fewer convergence iterations. / Computer and Information Science

Yes / The removal of pollutants such as N-nitrosamine present in drinking and reuse water resources is of significant interest for health and safety professionals. Reverse osmosis (RO) is one of the most promising and efficient methodologies for removing such harmful organic compounds from wastewater. Having said this, the literature confirms that the multi-stage RO process with retentate reprocessing design has not yet achieved an effective removal of N-nitrosodimethylamine-D6 (NDMA) from wastewater. This research emphasizes on this particular challenge and aims to explore several conceptual designs of multi-stage RO processes for NDMA rejection considering model-based techniques and compute the total recovery rate and energy consumption for different configurations of retentate reprocessing techniques. In this research, the permeate reprocessing design methodology is proposed to increase the process efficiency. An extensive simulation analysis is carried out using high NDMA concentration to evaluate the performance of each configuration under similar operational conditions, thus providing a deep insight on the performance of the multi-stage RO permeate reprocessing predictive design. Furthermore, an optimisation analysis is carried out on the final design to optimise the process with a high NDMA rejection performance and the practical recovery rate by manipulating the operating conditions of the plant within specified constraints bounds. The results show a superior removal of NDMA from wastewater.

Yes / Reverse Osmosis (RO) is one of the most widely used technologies for wastewater treatment for the removal of toxic impurities, such as phenol and phenolic compounds from industrial effluents. In this research, performance of multi-stage RO wastewater treatment system is evaluated for the removal of chlorophenol from wastewater using model-based techniques. A number of alternative configurations with recycling of permeate, retentate, and permeate-retentate streams are considered. The performance is measured in terms of total recovery rate, permeate product concentration, overall chlorophenol rejection and energy consumption and the effect of a number of operating parameters on the overall performance of the alternative configurations are evaluated. The results clearly show that the permeate recycling scheme at fixed plant feed flow rate can remarkably improve the final chlorophenol concentration of the product despite a reduction in the total recovery rate.

Yes / Multi-stage attacks (MSAs) are among the most serious threats in cyberspace today. Criminals target big organisations and government critical infrastructures mainly for financial gain. These attacks are becoming more advanced and stealthier, and thus have capabilities to evade Intrusion Detection Systems (IDSs). As a result, the attack strategies used in the attack render IDSs ineffective, particularly because of new security challenges introduced by some of the key emerging technologies such as 5G wireless networks, cloud computing infrastructure and Internet of Things (IoT), Advanced persistent threats (APTs) and botnet attacks are examples of MSAs, these are serious threats on the Internet. This work analyses recent MSAs, outlines and reveals open issues, challenges and opportunities with existing detection methods.

Recent years have seen more and more companies digitize their logistics systems to varying degrees. Data sharing standards for the supply chain have appeared in different fields, such as ONE Record for air cargo transport, papiNet for the paper and forest industry. One of the existing challenges is that it is difficult to carry out horizontal integration between companies from different supply chains because they each use different data exchange standards. Hence it is challenging to achieve wider-scale sustainability in this case. DigiGoods (a Vinnova funded project) focused on bringing improvements through digitization and data sharing through participants in the logistics value chain. It proposes a data model to build its data exchange standard between supply chain partners for sharing data and synchronizing progress. This thesis explores how to integrate this standard with other existing standards, and on this basis, explores how to use machine learning to optimize forecasts in the supply chain. This thesis lays the foundation for the integration of standards in the supply chain, explores the application of machine learning in the supply chain, and applies machine learning algorithms in multiple stages to improve the accuracy of forecasts in the supply chain, thereby responding to the global crisis. The performance of eight machine learning models is tested and compared to find the optimal algorithm and parameters for each dataset. A prototype is implemented to combine the advantages of the eight models and demonstrate that multi-stage machine learning could improve the prediction results in the context of DigiGoods. / De senaste åren har allt fler företag digitaliserat sina logistiksystem i varierande grad. Datadelningsstandarder för leveranskedjan har dykt upp på olika områden, till exempel ONERecord för flygfrakt, papiNet för pappers- och skogsindustrin. En av de befintliga utmaningarna är att det är svårt att genomföra horisontell integration mellan företag från olika leveranskedjor eftersom de använder olika datautbytesstandarder. om förbättringar genom digitalisering och datadelning genom deltagare i logistikvärdekedjan. Den föreslår en datamodell för att bygga sina standarder för datautbyte mellan leverantörskedjepartner för att dela data och synkronisera framsteg. Denna utforskar hur man integrerar denna standard med andra befintliga standarder, och på denna grund undersöker man hur man använder maskininlärning för att optimera prognoser i leveranskedjan. Denna avhandling lägger grunden för integration av standarder i försörjningskedjan, utforskar tillämpningen av maskininlärning i leveranskedjan och tillämpar maskininlärningsalgoritmer i flera steg för att förbättra prognosernas noggrannhet i leveranskedjan och därmed svara på den globala krisen. Prestanda för åtta maskininlärningsmodeller testas och jämförs för att hitta den optimala algoritmen och parametrarna för varje datamängd. Aprototyp implementeras för att kombinera fördelarna med de åtta modellerna och visa att maskininlärning i flera steg kan förbättra förutsägelseresultaten inom ramen för DigiGoods.

Recent years have seen more and more companies digitize their logistics systems to varying degrees. Data sharing standards for the supply chain have appeared in different fields, such as ONE Record for air cargo transport, papiNet for the paper and forest industry. One of the existing challenges is that it is difficult to carry out horizontal integration between companies from different supply chains because they each use different data exchange standards. Hence it is challenging to achieve wider-scale sustainability in this case. DigiGoods (a Vinnova funded project) focused on bringing improvements through digitization and data sharing through participants in the logistics value chain. It proposes a data model to build its data exchange standard between supply chain partners for sharing data and synchronizing progress. This thesis explores how to integrate this standard with other existing standards, and on this basis, explores how to use machine learning to optimize forecasts in the supply chain. This thesis lays the foundation for the integration of standards in the supply chain, explores the application of machine learning in the supply chain, and applies machine learning algorithms in multiple stages to improve the accuracy of forecasts in the supply chain, thereby responding to the global crisis. The performance of eight machine learning models is tested and compared to find the optimal algorithm and parameters for each dataset. A prototype is implemented to combine the advantages of the eight models and demonstrate that multi-stage machine learning could improve the prediction results in the context of DigiGoods. / De senaste åren har allt fler företag digitaliserat sina logistiksystem i varierande grad. Datadelningsstandarder för leveranskedjan har dykt upp på olika områden, till exempel ONE Record för flygfrakt, papiNet för pappers- och skogsindustrin. En av de befintliga utmaningarna är att det är svårt att genomföra horisontell integration mellan företag från olika leveranskedjor eftersom de använder olika datautbytesstandarder. om förbättringar genom digitalisering och datadelning genom deltagare i logistikvärdekedjan. Den föreslår en datamodell för att bygga sina standarder för datautbyte mellan leverantörskedjepartner för att dela data och synkronisera framsteg. Denna utforskar hur man integrerar denna standard med andra befintliga standarder, och på denna grund undersöker man hur man använder maskininlärning för att optimera prognoser i leveranskedjan. Denna avhandling lägger grunden för integration av standarder i försörjningskedjan, utforskar tillämpningen av maskininlärning i leveranskedjan och tillämpar maskininlärningsalgoritmer i flera steg för att förbättra prognosernas noggrannhet i leveranskedjan och därmed svara på den globala krisen. Prestanda för åtta maskininlärningsmodeller testas och jämförs för att hitta den optimala algoritmen och parametrarna för varje datamängd. Aprototyp implementeras för att kombinera fördelarna med de åtta modellerna och visa att maskininlärning i flera steg kan förbättra förutsägelseresultaten inom ramen för DigiGoods

In this dissertation, we study several stochastic capacity expansion models in the presence of permanent, spot market, and contract capacity for acquisition. Using a scenario tree approach to handle the data uncertainty of the problems, we develop multi-stage stochastic integer programming formulations for these models. First, we study multi-period single resource stochastic capacity expansion problems, where different sources of capacity are available to the decision maker. We develop efficient algorithms that can solve these models to optimality in polynomial time. Second, we study multi-period stochastic network capacity expansion problems with different sources for capacity. The proposed models are NP-hard multi-stage stochastic integer programs and we develop an efficient, asymptotically convergent approximation algorithm to solve them. Third, we consider some decomposition algorithms to solve the proposed multi-stage stochastic network capacity expansion problem. We propose an enhanced Benders' decomposition algorithm to solve the problem, and a Benders' decomposition-based heuristic algorithm to find tight bounds for it. Finally, we extend the stochastic network capacity expansion model by imposing budget restriction on permanent capacity acquisition cost. We design a Lagrangian relaxation algorithm to solve the model, including heuristic methods to find tight upper bounds for it. / Thesis / Doctor of Philosophy (PhD)