Multi-instance data, in which each object (e.g., a document) is a collection of instances (e.g., word), are widespread in machine learning, signal processing, computer vision, bioinformatic, music, and social sciences. Existing probabilistic models, e.g., latent Dirichlet allocation (LDA), probabilistic latent semantic indexing (pLSI), and discrete component analysis (DCA), have been developed for modeling and analyzing multiinstance data. Such models introduce a generative process for multi-instance data which includes a low dimensional latent structure. While such models offer a great freedom in capturing the natural structure in the data, their inference may present challenges. For example, the sensitivity in choosing the hyper-parameters in such models, requires careful inference (e.g., through cross-validation) which results in large computational complexity. The inference for fully Bayesian models which contain no hyper-parameters often involves slowly converging sampling methods. In this work, we develop approaches for addressing such challenges and further enhancing the utility of such models. This dissertation demonstrates a unified convex framework for probabilistic modeling of multi-instance data. The three main aspects of the proposed framework are as follows. First, joint regularization is incorporated into multiple density estimation to simultaneously learn the structure of the distribution space and infer each distribution. Second, a novel confidence constraints framework is used to facilitate a tuning-free approach to control the amount of regularization required for the joint multiple density estimation with theoretical guarantees on correct structure recovery. Third, we formulate the problem using a convex framework and propose efficient optimization algorithms to solve it. This work addresses the unique challenges associated with both discrete and continuous domains. In the discrete domain we propose a confidence-constrained rank minimization (CRM) to recover the exact number of topics in topic models with theoretical guarantees on recovery probability and mean squared error of the estimation. We provide a computationally efficient optimization algorithm for the problem to further the applicability of the proposed framework to large real world datasets. In the continuous domain, we propose to use the maximum entropy (MaxEnt) framework for multi-instance datasets. In this approach, bags of instances are represented as distributions using the principle of MaxEnt. We learn basis functions which span the space of distributions for jointly regularized density estimation. The basis functions are analogous to topics in a topic model. We validate the efficiency of the proposed framework in the discrete and continuous domains by extensive set of experiments on synthetic datasets as well as on real world image and text datasets and compare the results with state-of-the-art algorithms. / Graduation date: 2013

Physical restraint use and falls are common in institutional care of old people and various attempts have been made to reduce their occurrence. Falls and concomitant injuries are a major problem due to their negative effect on morbidity and mortality. Prevention of falls and injuries is the most common reason for physically restraining old people in institutional care. Its use has, however, been questioned both from an ethical perspective, since restraints can be perceived as coercive and also because of the lack of sound evidence of their effectiveness in preventing falls, as well as the adverse effects associated with their use. The main purposes of this thesis were to investigate differences in the us of physical restraints over time, to identifify risk factors for falls among people with dementia, to evaluate the effects of a restraint minimization program on staff knowledge, attitudes, and work environment and use of physical restraints and the quality of care. The present thesis is based on three main data collections, two census surveys conducted within institutional care for old people in the county of Västerbotten in 2000 (n=3,804) and 2007 (n=2,970) and one cluster-Randomized Controlled Trial (RCT) including 40 group dwellings for people suffering from dementia where the intervention consisted of staff education. The use of physical restraints increased slightly between 2000 and 2007 (16.2% to 18.4%, p=0.016). Analyses suggest that the increase might be independent of any change in resident characteristics. Restrained residents were also subjected to restraints for longer times in 2007. During a six-month follow-up 64/160 (40.0%) residents in group dwellings for those with dementia sustained at least one fall. Independent risk factors for falls were ‘requiring help with hygiene’, ‘displaying verbally disruptive/attention-seeking behavior’, ‘able to rise from a chair’, ‘walking with assistive devices’, and ‘participating in outdoor walks’, which explained 36.1% of the falls. The majority of the 191 falls were un-witnessed, 35% occurred during the night and anxiety and confusion were the most common symptoms preceding the falls. A six-month restraint minimization program showed a positive impact on staff knowledge, attitudes and work environment as well as on the use of physical restraints and subjectively estimated quality of care. Residents in the intervention group present throughout the entire study period had lower odds, relative to the residents in the control group of being physically restrained at follow-up (OR= 0.21, CI 95%=0.08-0.57) after controlling for potential confounders and the cluster effect. Adjusted analyses including all residents present at either baseline or follow-up also showed that the use of physical restraints was less in the intervention group relative to the control group at follow-up. There was no change in the occurrence of falls or use of psychoactive drugs. The intervention also reduced stress of conscience, job demands and strain in the staff, and improved their job control and the caring climate. Subgroup analysis indicated a greater effect in units where the use of physical restraints had been reduced or remained constant. In conclusion, physical restraint use and falls remains common in institutional care of old people. The practice of physical restraint seems to have changed. In the RCT it was found that it is possible to change restraint practice and also to improve staff work environment. Falls among residents with dementia require a certain mobility function and anxiety and confusion are common symptoms preceding falls.

This thesis (1) reports the development of simulation tools and control strategies for optimizing hybrid electric vehicle (HEV) energy management, and (2) reports the design and testing of a hydraulic hybrid school bus (HHB) prototype. A hybrid vehicle is one that combines two or more energy sources for use in vehicle propulsion. Hybrid electric vehicles have become popular in the consumer market due to their greatly improved fuel economy over conventional vehicles. The control strategy of an HEV has a paramount effect on its fuel economy performance. In this thesis, backward-looking and forward-looking simulations of three HEV architectures (parallel, power-split and 2-mode power-split) are developed. The Equivalent Cost Minimization Strategy (ECMS), which weights electrical power as an equivalent fuel usage, is then studied in great detail and improvements are suggested. Specifically, the robustness of an ECMS controller is improved by linking the equivalence factor to dynamic programming and then further tailoring its functional form. High-fidelity vehicle simulations over multiple drive-cycles are performed to measure the improved performance of the new ECMS controller, and to show its potential for online application. While HEVs are prominent in the consumer market and studied extensively in current literature, hydraulic hybrid vehicles (HHVs) only exist as heavy utility vehicle prototypes. The second half of this thesis reports design, construction, and testing of a hydraulic hybrid school bus prototype. Design considerations, simulation results, and preliminary testing results are reported, which indicate the strong potential for hydraulic hybrids to improve fuel economy in the school bus vehicle segment.

In this paper we propose an algorithm for solving systems of coupled monotone inclusions in Hilbert spaces. The operators arising in each of the inclusions of the system are processed in each iteration separately, namely, the single-valued are evaluated explicitly (forward steps), while the set-valued ones via their resolvents (backward steps). In addition, most of the steps in the iterative scheme can be executed simultaneously, this making the method applicable to a variety of convex minimization problems. The numerical performances of the proposed splitting algorithm are emphasized through applications in average consensus on colored networks and image classification via support vector machines.

This thesis is targeted at precoding methods and resource allocation for the downlink of fixed multi-user multi-antenna broadband wireless systems. We explore different utilizations of precoders in transmission over frequency-selective channels. We first consider the weighted sum-rate (WSR) maximization problem for multi-carrier systems using linear precoding and propose a low complexity algorithm which exhibits near-optimal performance. Moreover, we offer a novel rate allocation method that utilizes the signalto- noise-ratio (SNR) gap to capacity concept to choose the rates to allocate to each data stream. We then study a single-carrier transmission scheme that overcomes known impairments associated with multi-carrier systems. The proposed scheme utilizes timereversal space-time block coding (TR-STBC) to orthogonalize the downlink receivers and performs the required pre-equalization using Tomlinson-Harashima precoding (THP).We finally discuss the strengths and weaknesses of the proposed method.

This thesis is targeted at precoding methods and resource allocation for the downlink of fixed multi-user multi-antenna broadband wireless systems. We explore different utilizations of precoders in transmission over frequency-selective channels. We first consider the weighted sum-rate (WSR) maximization problem for multi-carrier systems using linear precoding and propose a low complexity algorithm which exhibits near-optimal performance. Moreover, we offer a novel rate allocation method that utilizes the signalto- noise-ratio (SNR) gap to capacity concept to choose the rates to allocate to each data stream. We then study a single-carrier transmission scheme that overcomes known impairments associated with multi-carrier systems. The proposed scheme utilizes timereversal space-time block coding (TR-STBC) to orthogonalize the downlink receivers and performs the required pre-equalization using Tomlinson-Harashima precoding (THP).We finally discuss the strengths and weaknesses of the proposed method.

University of Macau / Faculty of Science and Technology / Department of Civil and Environmental Engineering

Closed-loop supply chains (CLSCs) that integrate the activities for reclaiming residual values in postconsumer products with the traditional forward supply chain activities are important from financial and environmental perspectives. This thesis develops models and analyses on three topics novel to the field of CLSC research with a goal of advancing knowledge about effective decision-makings in CLSCs. In the first part of the thesis, we study joint control of stochastic forward and stochastic reverse material flows in CLSCs. With an application to a CLSC where postconsumer products are collected for warranty service purposes, we demonstrate that the benefit of coordinating two production activities could be significant. We develop a model that can be used to obtain an effective inventory control policy for coordinating forward and reverse material flows. Through Monte Carlo simulation and global sensitivity analysis, we identify major influential factors that affect system's warranty cost savings performance. The results indicate that joint control of forward and reverse material flows greatly improves warranty cost savings performance as well as system's robustness to uncertainties. The second part of the thesis develops a differential game model for characterizing decentralized time-varying competitive decision-making in a CLSC. The differential game model is particularly useful for studying time-varying interactive decision-making in CLSCs that involve many stakeholders who pursue different objectives in forward and reverse production activities. We identify optimal prices and production strategies that evolve over time under fluctuating market demand. Also, the model provides a quantitative scheme that can be used to obtain an efficient apportionment of product recovery processes. The third part of the thesis describes the relationship among consumers' risk-aversion, product cannibalization of new products by remanufactured products, and growth of CLSCs through price optimization models. Whereas price is one of the most effective variables for managing market demand, previous CLSC research has mainly focused on operational problems without paying much attention on the interface between CLSCs and markets. We develop models that jointly determine optimal prices in forward and reverse channels considering consumers' willingness-to-pay (WTP) for remanufactured products, consumers' willingness-to-accept (WTA) for a buyback price, and consumers' risk aversion to uncertain quality perceptions. The results show that consumers' active participation in CLSC is an important factor for the viability and growth of a CLSC. Also, we show that companies can benefit from product remanufacturing although it may be accompanied by production cannibalization.

Scene classification is performed on countless remote sensing images in support of operational activities. Automating this process is preferable since manual pixel-level classification is not feasible for large scenes. However, developing such an algorithmic solution is a challenging task due to both scene complexities and sensor limitations. The objective is to develop efficient and accurate unsupervised methods for classification (i.e., assigning each pixel to an appropriate generic class) and for labeling (i.e., properly assigning true labels to each class). Unique from traditional approaches, the proposed bootstrap approach achieves classification and labeling without training data. Here, the full image is partitioned into subimages and the true classes found in each subimage are provided by the user. After these steps, the rest of the process is automatic. Each subimage is individually classified into regions and then using the joint information from all subimages and regions the optimal configuration of labels is found based on an objective function based on a Markov random field (MRF) model. The bootstrap approach has been successfully demonstrated with SAR sea-ice and lake ice images which represent challenging scenes used operationally for ship navigation, climate study, and ice fraction estimation. Accuracy assessment is based on evaluation conducted by third party experts. The bootstrap method is also demonstrated using synthetic and natural images. The impact of this technique is a repeatable and accurate methodology that generates classified maps faster than the standard methodology.

In an ever-changing global environment, the ability to adapt to the current economic climate is essential for a company to prosper and survive. Numerous previous re- search state that better risk management and low overall risks will lead to a higher firm value. The purpose of this study is to examine if portfolio theory, made for fi- nancial portfolios, can be used to compose product portfolios in order to minimize risk and optimize returns. The term contingent hedge is defined as an optimal portfolio that can be identified today, that in the future will yield a stable stream of returns at a low level of risk. For companies that might engage in costly hedging activities on the futures market, the benefits of creat- ing a contingent hedge are several. These include creating an optimized portfolio that minimizes risk and avoid trading contracts on futures markets that would incur hefty transaction costs and risks. Using quantitative financial models, product portfolio compositions are generated and compared with the returns and risks profile of individual commodities, as well as the actual product portfolio compositions of publicly traded mining companies. Us- ing Modern Portfolio Theory an efficient frontier is generated, yielding two inde- pendent portfolios, the minimum risk portfolio and the tangency portfolio. The Black-Litterman model is also used to generate yet another portfolio using a Bayesian approach. The portfolios are generated by historic time-series data and compared with the actual future development of commodities; the portfolios are then analyzed and compared. The results indicate that the minimum risk portfolio provides a signif- icantly lower risk than the compositions of all mining companies in the study, as well as the risks of individual commodities. This in turn will lead to several benefits for company management and the firm’s shareholders that are discussed throughout the study. However, as for a return-optimizing portfolio, no significant results can be found. Furthermore, the analysis suggests a series of improvements that could potentially yield an even greater result. The recommendation is that mining companies can use the methods discussed throughout this study as a way to generate a costless contin- gent hedge, rather than engage in hedging activities on futures markets.