The classification of data with imbalanced class distributions has posed a significant drawback in the performance attainable by most well-developed classification systems, which assume relatively balanced class distributions. This problem is especially crucial in many application domains, such as medical diagnosis, fraud detection, network intrusion, etc., which are of great importance in machine learning and data mining. This thesis explores meta-techniques which are applicable to most classifier learning algorithms, with the aim to advance the classification of imbalanced data. Boosting is a powerful meta-technique to learn an ensemble of weak models with a promise of improving the classification accuracy. AdaBoost has been taken as the most successful boosting algorithm. This thesis starts with applying AdaBoost to an associative classifier for both learning time reduction and accuracy improvement. However, the promise of accuracy improvement is trivial in the context of the class imbalance problem, where accuracy is less meaningful. The insight gained from a comprehensive analysis on the boosting strategy of AdaBoost leads to the investigation of cost-sensitive boosting algorithms, which are developed by introducing cost items into the learning framework of AdaBoost. The cost items are used to denote the uneven identification importance among classes, such that the boosting strategies can intentionally bias the learning towards classes associated with higher identification importance and eventually improve the identification performance on them. Given an application domain, cost values with respect to different types of samples are usually unavailable for applying the proposed cost-sensitive boosting algorithms. To set up the effective cost values, empirical methods are used for bi-class applications and heuristic searching of the Genetic Algorithm is employed for multi-class applications. This thesis also covers the implementation of the proposed cost-sensitive boosting algorithms. It ends with a discussion on the experimental results of classification of real-world imbalanced data. Compared with existing algorithms, the new algorithms this thesis presents are superior in achieving better measurements regarding the learning objectives.

The classification of data with imbalanced class distributions has posed a significant drawback in the performance attainable by most well-developed classification systems, which assume relatively balanced class distributions. This problem is especially crucial in many application domains, such as medical diagnosis, fraud detection, network intrusion, etc., which are of great importance in machine learning and data mining. This thesis explores meta-techniques which are applicable to most classifier learning algorithms, with the aim to advance the classification of imbalanced data. Boosting is a powerful meta-technique to learn an ensemble of weak models with a promise of improving the classification accuracy. AdaBoost has been taken as the most successful boosting algorithm. This thesis starts with applying AdaBoost to an associative classifier for both learning time reduction and accuracy improvement. However, the promise of accuracy improvement is trivial in the context of the class imbalance problem, where accuracy is less meaningful. The insight gained from a comprehensive analysis on the boosting strategy of AdaBoost leads to the investigation of cost-sensitive boosting algorithms, which are developed by introducing cost items into the learning framework of AdaBoost. The cost items are used to denote the uneven identification importance among classes, such that the boosting strategies can intentionally bias the learning towards classes associated with higher identification importance and eventually improve the identification performance on them. Given an application domain, cost values with respect to different types of samples are usually unavailable for applying the proposed cost-sensitive boosting algorithms. To set up the effective cost values, empirical methods are used for bi-class applications and heuristic searching of the Genetic Algorithm is employed for multi-class applications. This thesis also covers the implementation of the proposed cost-sensitive boosting algorithms. It ends with a discussion on the experimental results of classification of real-world imbalanced data. Compared with existing algorithms, the new algorithms this thesis presents are superior in achieving better measurements regarding the learning objectives.

Research on self-assembling gemini surfactants and other amphiphiles for potential gene delivery applications in research as well as in clinical practice, and as alternatives to viral gene delivery vectors, is beginning to focus more on structureactivity relationships to address the current low gene delivery efficiencies of amphiphiles. Some underlying structureactivity relations are beginning to emerge. But, as a better understanding of the factors that govern the transfection abilities of amphiphile molecules emerges, development of improved non-viral vectors with clinical potential may also emerge.<p> The research conducted for this thesis was aimed at the design, synthesis and in vitro investigation of gemini surfactants as one of a family of novel amphiphiles being investigated for gene therapeutic applications. The properties of these compounds can be controlled as well as allowed to vary naturally. Gemini surfactant-based gene delivery systems were prepared and characterized for transfer of Luciferase plasmid (pMASIA.Luc) to both COS-7 and PAM 212 cells. Characterization was accomplished using microscopy, dynamic light scattering (DLS) and zeta (ζ) potential analysis. In vitro gene expression and toxicities were evaluated in COS-7 cell and PAM 212 keratinocyte cultures.<p> The level of in vitro transfection in general was found to correlate strongly with the structure of the gemini surfactants. Among the 12-spacer-12 surfactants, incorporation of a pH-sensitive aza (N-CH3) group, which is also steric hindrance-imposing, in the spacer chain yielded increased transfection, particularly for the 12-7N-12 surfactant. In comparison, the incorporation of the more pH-sensitive imino (N-H) group in the 12-7NH-12 surfactant yielded the highest increase in transfection among the 12-spacer-12 surfactants. The deleterious effect of steric hindrance due to the aza group is more evident when comparing the transfection efficiency of 12-5N-12 (1 × aza, higher) vs. 12-8N-12 (2 × aza, lower transfection). Another highlighted structural feature is provided by the fact that both the 12-7NH-12 and 12-7N-12 surfactants had higher transfection efficiencies than 12-5N-12 and 12-8N-12 surfactants; the first pair has trimethylene spacing, which constitutes an optimal separation between nitrogen centres, while the second pair has shorter dimethylene spacings.<p> After expanding the structure of surfactants, transfection efficiencies were found to increase in response to increase in hydrocarbon tail length, but were much lower for surfactants with no amino functional groups, those that lacked the optimal trimethylene spacing, or those having both of these limitations in the gemini surfactant spacer. The 18-7NH-18 surfactant had the highest overall transfection in both COS-7 and PAM 212 cells. Gemini surfactant-based gene delivery systems capable of adopting both polymorphic structural phases and which could undergo pH-induced structural transition demonstrated high transfection efficiencies. Gemini surfactants with both characteristics (e.g., 12-7NH-12-based complexes are both polymorphic and pH-sensitive) had higher transfection than gemini surfactants with only one (e.g., 12-3-12-based complexes are only polymorphic).<p> Overall, the m-7NH-m surfactants, the most efficient surfactants studied, had transfection efficiencies similar to that of the commercial Lipofectamine Plus reagent and imposed no higher toxicity on cells relative to the less efficient surfactants. Thus, the design of the m-7NH-m surfactants to enhance their transfection abilities also ensured that their toxicity to cells were kept minimal. Overall, the design, synthesis and in vitro transfection screening of gemini surfactant candidates has revealed that the m-7NH-m surfactants have the highest transfection efficiencies; they have emerged as suitable candidates for non-viral gene delivery in vivo or at higher levels. Gene delivery investigations for six of the gemini surfactant candidates are being reported for the first time.

Soil salinity affects over 20% of the worlds irrigated land. Potato (Solanum tuberosum L.), the most important vegetable crop worldwide, is relatively salt sensitive. However, relatively little work has been done on salt tolerance of the potato plant. This thesis investigated the methodology of treatment application and scion/rootstock effects on subsequent salt stress responses of four contrasting potato genotypes: Norland, 9506, 9120-05 [ABA-deficient mutant], and 9120-18 [ABA-normal sibling] grown hydroponically in sand. The effect of incremental salt stress were studied, using NaCl, CaCl2 and combined NaCl + CaCl2 pre- treatments as well as varying methods of ABA application with a specific focus on the role of rootstock and scion. Physiological responses of various potato genotypes to salt stress differed depending on how the salt stress was applied. An incremental salt stress regime was able to more effectively differentiate genotypes based on salt stress resistance and greater salt tolerance compared to a sudden salt shock. Generally, the ability to produce ABA was positively related to the degree of salt stress resistance, with higher ABA levels induced under incremental salt stress treatments compared to salt shock. The method of ABA application also had a marked effect on potato responses to salt stress. Slowly increasing concentrations of exogenous ABA maintained growth rates, enhanced root water content and induced more lateral shoot growth compared to a single ABA dose. The degree of salt tolerance induced by the grafted rootstock was primarily modulated by salt acclimation and was manifested in the scion as increased water content, stem diameter, dry matter accumulation, stomatal conductivity, and osmotic potential and was associated with reduced leaf necrosis. Using the salt-resistant 9506 line as a scion also significantly increased root fresh and dry weight and stem diameter as well as root water content of salt-sensitive ABA-deficient mutant rootstocks. Exogenous ABA appeared to enhance plant water status via the roots under salt stress beyond that of grafting alone. This was verified by more positive stomatal conductivity and greater upward water flow in ABA treated grafted and non-grafted plants as compared to the absence of upward water flow in non-treated grafted plants as measured via micro NMR imaging. NaCl pre-treatment produced greater salt stress resistance compared to pre-treatment with CaCl2 and was associated with a specific Na+ ion effect rather than a non-specific EC-dependent response. However, the presence of both ABA and CaCl2 appears to be necessary in order to enhance Na+ exclusion from the shoot and increases the K+/Na+ ratio.

Beta-amyloid (Aâ) remains to be the focus of research interest of the pathogenesis of Alzheimers disease (AD). Aâ is subject to oligomerization and its polymers are cytotoxic. Advanced aggregation leads to formation of senile plaques. Depositions of Aâ surrounding the cerebral vasculature, i.e. cerebral amyloid angiopathy (CAA), occur in most AD patients. The occurrence of Aâ aggregation in AD brains is not due to over-expression of amyloid precursor protein in most cases of AD. Factors influencing Aâ polymerization are yet to be established.<p> Aldehydes are highly reactive. They can cause protein crosslinkage. It is interesting to study whether endogenous aldehydes may be involved in Aâ polymerization process. In order to investigate the potential interaction of endogenous aldehydes with Aâ and their effects on its aggregation, various techniques including thioflavin T fluometry, dynamic light scattering, circular dichroism and atomic force microscopy were employed to assess Aâ aggregation at different stages. Formaldehyde, methylglyoxal, malondialdehyde and 4-hydroxyl-nonenal were found to enhance Aâ â-sheets formation, oligomerization and fibrillogenesis in vitro. The sizes of the oligomers are increased after interaction with the aldehydes. Lysine residues of Aâ were identified to be the primary site of interaction with aldehydes by forming Schiff bases, which may subsequently lead to intra- and inter-molecular crosslinkage. Aldehydes can also crosslink Aâ with other proteins such as apolipoprotein E and á2-macroglobulin (á2M), to form large complexes. Results suggest that aldehydes substantially increase the rate of Aâ oligomerization at each stage of fibrillogenesis.<p> The native and formaldehyde-modified Aâ oligomers were isolated by size exclusion chromatography and their cytotoxic effects towards SH-SY5Y neuroblastoma cells were assessed using MTT, LDH and caspase-3 activity assays. The aldehyde-modified oligomers are slightly but significantly more cytotoxic compared to the native oligomers. Since aldehydes significantly increase the production of Aâ oligomers, an increase in aldehydes would enhance the total cytotoxicity, suggesting that aldehydes may potentially exacerbate neurovascular damage and neurodegeneration caused by Aâ.<p> Low-density lipoprotein receptor related protein-1 (LRP-1) plays a crucial role in Aâ clearance via the cerebral vasculature. Semicarbazide-sensitive amine oxidase (SSAO) and LRP-1 are both richly expressed on the vascular smooth muscle cells (VSMCs). We demonstrated that SSAO-mediated deamination affects LRP-1 function using isolated VSMCs. Formaldehyde at low concentrations decreases LRP-1-mediated uptake of á2M, a substrate of LRP-1 and a carrier for Aâ. Methylamine, an SSAO substrate that is converted to formaldehyde, also inactivates LRP-1 function, but not in the presence of an SSAO inhibitor. Increased SSAO-mediated deamination can potentially impair Aâ clearance via LRP-1.<p> In conclusion, aldehydes derived from oxidative stress and SSAO-mediated deamination induce Aâ aggregation, enhance Aâ cytotoxicity and impair Aâ clearance. The exclusive localization of SSAO on the cerebral vasculature may be responsible for the perivascular deposition of Aâ, i.e. CAA, which is associated both with vascular dementia and with AD. Vascular surface SSAO may be a novel pharmacological target for the treatment of AD.

<p>The visual system is thought to represent the trajectory of moving objects in the activity of large populations of cortical neurons that respond preferentially to the direction of stimulus motion. Here I employed in vivo voltage sensitive dye (VSD) imaging to explore how abrupt changes in the trajectory of a moving stimulus impact the population coding of motion direction in ferret primary visual cortex (V1). For motion in a constant direction, the peak of the cortical population response reliably signaled the stimulus trajectory; but for abrupt changes in motion direction, the peak of the population response departed significantly from the stimulus trajectory in a fashion that depended on the size of the direction deviation. For small direction deviation angles, the peak of the active population shifted from values consistent with the initial direction of motion to those consistent with the final direction of motion by progressing smoothly through intermediate directions not present in the stimulus. In contrast, for large direction deviation angles, peak values consistent with the initial motion direction were followed by: a small deviation away from the final motion direction, a rapid 180° jump, and a gradual shift to the final direction. These departures of the population response from the actual trajectory of the stimulus predict specific misperceptions of motion direction that were confirmed by human psychophysical experiments. I conclude that cortical dynamics and population coding mechanisms combine to place constraints on the accuracy with which abrupt changes in direction of motion can be represented by cortical circuits.</p> / Dissertation

Risk attitudes are important for human decision making, especially in scenarios where huge wins or losses are possible, as exemplified by planetary rover navigation, oilspill response, and business applications. Decision-theoretic planners therefore need to take risk aspects into account to serve their users better. However, most existing decision-theoretic planners use simplistic planning objectives that are risk-neutral. The thesis research is the first comprehensive study of how to incorporate risk attitudes into decision-theoretic planners and solve large-scale planning problems represented as Markov decision process models. The thesis consists of three parts. The first part of the thesis work studies risk-sensitive planning in case where exponential utility functions are used to model risk attitudes. I show that existing decision-theoretic planners can be transformed to take risk attitudes into account. Moreover, different versions of the transformation are needed if the transition probabilities are implicitly given, namely, temporally extended probabilities and probabilities given in a factored form. The second part of the thesis work studies risk-sensitive planning in case where general nonlinear utility functions are used to model risk attitudes. I show that a state-augmentation approach can be used to reduce a risk-sensitive planning problem to a risk-neutral planning problem with an augmented state space. I further use a functional interpretation of value functions and approximation methods to solve the planning problems efficiently with value iteration. I also show an exact method for solving risk-sensitive planning problems where one-switch utility functions are used to model risk attitudes. The third part of the thesis work studies risk sensitive planning in case where arbitrary rewards are used. I propose a spectrum of conditions that can be used to constrain the utility function and the planning problem so that the optimal expected utilities exist and are finite. I prove that the existence and finiteness properties hold for stationary plans, where the action to perform in each state does not change over time, under different sets of conditions.

The detection of volatile organic compounds in liquid is of interest for applications in public health, workplace safety and environmental monitoring. Traditionally, water samples were taken and analyzed in the laboratory using classical laboratory instrumentation. Current trends target real-time measurements using e.g. chemical microsensors built with microfabrication technologies. Among these, mass-sensitive chemical sensors, based on cantilever beams or surface acoustic devices, have shown substantial promise in gas-phase applications. In a liquid environment, the resonant microstructures typically suffer from high damping, which negatively affects the sensor resolution. In this work, a novel disk-type resonator developed at Georgia Tech was investigated as chemical microsensor for liquid-phase applications. The micromachined resonator vibrates in a rotational in-plane mode shape, reducing damping in a liquid environment. As part of the present research, a measurement setup with a custom-made flow cell for liquid-phase chemical measurements and a coating system to locally deposit polymer sensitive films onto the resonators were developed. To improve the film adhesion on the resonator surface in liquid, physical and chemical binding techniques were developed and tested on wafer samples. Polymers such as poly(4-vinylpyrrolidone), poly(ethylene-co-propylene) and poly(styrene-co-butadiene) were deposited using the custom-designed coating system onto the disk-type resonators. Liquid-phase measurements using tetrachloroethylene as the chemical analyte were performed. The experimental results are discussed, sources of problems are identified and recommendations for future research are made.

Taiwan has an ageing population, with more people in the concern of not having sufficient income stream during retired life. This study investigates the ageing issue through socioeconomic perspectives. It is recommended that apart from low interest-bearing term deposits, Interest-Sensitive Annuity is the most suitable solution for countering longevity risk. Through case studies, it has been found that 1, due to Annuity Puzzle sentiment, term depositors will continue to invest in Interest-Sensitive Annuities, regardless of the low interest rate environment. 2, Interest-Sensitive Annuity investors are as risk-averse as term depositors, implying that they do not necessarily choose the surrender option upon expiry. 3, due to customer sentiment, the Interest-Sensitive Annuity policy fees charged are inversely correlated to customers¡¦ willingness to invest. 4, by selling low-commission products, namely one- and two-year Interest-Sensitive Annuities through bancassurance channel, insurance companies enjoy the benefit of low cost capital and are able to reduce interest spread risk through efficient investments. Moreover, customers have their retirement needs covered while insurance salespeople of different channels are able to meet respective sales targets. It is therefore shown that Interest-Sensitive Annuities have the following benefits. For investors, it is the product type that best meets their needs. For insurance salespeople, they enjoy a diverse and complete product portfolio and for insurance companies, it maximizes operation efficiency. Unfortunately, after the termination of one- and two-year Interest-Sensitive Annuities on the market, insurance company capital costs have been negatively impacted, with retirement and longevity risks unsatisfied and insurance salespeople having less products to choose from. It is suggested that the regulator considers re-introducing one- and two-year Interest-Sensitive Annuities, using Risk-Based Capital as a complement in monitoring insurance companies.

Based on many studies and experiments, blood flow patterns are associated with cardiovascular diseases and it usually is a sign of cardiovascular disease when the blood flow becomes unusual. Magnetic resonance (MR) imaging is a non-invasive medical technique and the characteristic of phase contrast can use to measure the flow velocity and patterns in vivo. As far as we know, for the cardiovascular of region of interest, 4D Flow-sensitive MRI technology is good at spatial coverage and temporal resolutions.It is easier for the researcher to analyze blood flow patterns in the clinical diagnosis by visualization processing. EnSight (CEI, USA), a kind of commercial software, is often used to do visualization processing of data of 4D Flow-sensitive MRI. However, before visualization of the data, several actions must be completed, such as ROI selection, correction or conversion of data, and etc. Therefore, our thesis hopes to develop a simple but practical user interface tool for 4D Flow-sensitive MRI data. From reading the file and ROI selection to correction and conversion of data as well as the visualization processing are completed by this tool. It provides researchers a rapid examination of data and analysis of blood flow patterns in the diagnosis.