Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

Relation between the level of self-mutilation and theconcentration of fecal metabolites of glucocorticoids incaptive chimpanzees (Pan troglodytes)

The influence of stress in an environment, according with the behavioral and endocrine variables of primates, are increasingly being studied by a diversity of authors, and have shown that abnormal behaviors associated with increased glucocorticoids may be directly related with the impairment of their well-being. In this work were used 22 adult chimpanzees (Pan troglodytes), 11 males and 11 females, kept in captivity in three different institutions. All animals had their behavior registered by focal session using a 30 seconds sample interval, during six months, totaling 4,800 registries per each animal. During this period, fecal samples were collected 3 times a week for the extraction and measurement of the concentration of fecal metabolites of glucocorticoid by radioimmunoassay. Of the total observed, stereotypical behaviors represented 13,45±2.76%, and among them, self-mutilation represented 38.28±3.98 %. The animals were classified into three different scores, according with the percentage of body surface with alopecia due to self-mutilation. It was found a positive correlation of high intensity between the scores of alopecia due to the observed mutilation and the average concentrations of fecal metabolites of glucocorticoids. This result strongly suggests that this measurement of self-mutilation in a chimpanzee can be used as an important auxiliary tool to evaluate de conditions of adaptation of an animal in captivity, functioning as a direct indicator of the presence of chronic stress.

Real-time Image-based RGB-D Camera Motion Tracking and Environment Mapping

In this work, image based estimation methods, also known as direct methods, are studied which avoid feature extraction and matching completely. Cost functions use raw pixels as measurements and the goal is to produce precise 3D pose and structure estimates. The cost functions presented minimize the sensor error, because measurements are not transformed or modified. In photometric camera pose estimation, 3D rotation and translation parameters are estimated by minimizing a sequence of image based cost functions, which are non-linear due to perspective projection and lens distortion. In image based structure refinement, on the other hand, 3D structure is refined using a number of additional views and an image based cost metric. Image based estimation methods are particularly useful in conditions where the Lambertian assumption holds, and the 3D points have constant color despite viewing angle. The goal is to improve image based estimation methods, and to produce computationally efficient methods which can be accomodated into real-time applications. The developed image-based 3D pose and structure estimation methods are finally demonstrated in practise in indoor 3D reconstruction use, and in a live augmented reality application.

Hydrometallurgical recovery of valuable metals from secondary raw materials

Wastes and side streams in the mining industry and different anthropogenic wastes often contain valuable metals in such concentrations their recovery may be economically viable. These raw materials are collectively called secondary raw materials. The recovery of metals from these materials is also environmentally favorable, since many of the metals, for example heavy metals, are hazardous to the environment. This has been noticed in legislative bodies, and strict regulations for handling both mining and anthropogenic wastes have been developed, mainly in the last decade. In the mining and metallurgy industry, important secondary raw materials include, for example, steelmaking dusts (recoverable metals e.g. Zn and Mo), zinc plant residues (Ag, Au, Ga, Ge, In) and waste slurry from Bayer process alumina production (Ga, REE, Ti, V). From anthropogenic wastes, waste electrical and electronic equipment (WEEE), among them LCD screens and fluorescent lamps, are clearly the most important from a metals recovery point of view. Metals that are commonly recovered from WEEE include, for example, Ag, Au, Cu, Pd and Pt. In LCD screens indium, and in fluorescent lamps, REEs, are possible target metals. Hydrometallurgical processing routes are highly suitable for the treatment of complex and/or low grade raw materials, as secondary raw materials often are. These solid or liquid raw materials often contain large amounts of base metals, for example. Thus, in order to recover valuable metals, with small concentrations, highly selective separation methods, such as hydrometallurgical routes, are needed. In addition, hydrometallurgical processes are also seen as more environmental friendly, and they have lower energy consumption, when compared to pyrometallurgical processes. In this thesis, solvent extraction and ion exchange are the most important hydrometallurgical separation methods studied. Solvent extraction is a mainstream unit operation in the metallurgical industry for all kinds of metals, but for ion exchange, practical applications are not as widespread. However, ion exchange is known to be particularly suitable for dilute feed solutions and complex separation tasks, which makes it a viable option, especially for processing secondary raw materials. Recovering valuable metals was studied with five different raw materials, which included liquid and solid side streams from metallurgical industries and WEEE. Recovery of high purity (99.7%) In, from LCD screens, was achieved by leaching with H2SO4, extracting In and Sn to D2EHPA, and selectively stripping In to HCl. In was also concentrated in the solvent extraction stage from 44 mg/L to 6.5 g/L. Ge was recovered as a side product from two different base metal process liquors with Nmethylglucamine functional chelating ion exchange resin (IRA-743). Based on equilibrium and dynamic modeling, a mechanism for this moderately complex adsorption process was suggested. Eu and Y were leached with high yields (91 and 83%) by 2 M H2SO4 from a fluorescent lamp precipitate of waste treatment plant. The waste also contained significant amounts of other REEs such as Gd and Tb, but these were not leached with common mineral acids in ambient conditions. Zn was selectively leached over Fe from steelmaking dusts with a controlled acidic leaching method, in which the pH did not go below, but was held close as possible to, 3. Mo was also present in the other studied dust, and was leached with pure water more effectively than with the acidic methods. Good yield and selectivity in the solvent extraction of Zn was achieved by D2EHPA. However, Fe needs to be eliminated in advance, either by the controlled leaching method or, for example, by precipitation. 100% Pure Mo/Cr product was achieved with quaternary ammonium salt (Aliquat 336) directly from the water leachate, without pH adjustment (pH 13.7). A Mo/Cr mixture was also obtained from H2SO4 leachates with hydroxyoxime LIX 84-I and trioctylamine (TOA), but the purities were 70% at most. However with Aliquat 336, again an over 99% pure mixture was obtained. High selectivity for Mo over Cr was not achieved with any of the studied reagents. Ag-NaCl solution was purified from divalent impurity metals by aminomethylphosphonium functional Lewatit TP-260 ion exchange resin. A novel preconditioning method, named controlled partial neutralization, with conjugate bases of weak organic acids, was used to control the pH in the column to avoid capacity losses or precipitations. Counter-current SMB was shown to be a better process configuration than either batch column operation or the cross-current operation conventionally used in the metallurgical industry. The raw materials used in this thesis were also evaluated from an economic point of view, and the precipitate from a waste fluorescent lamp treatment process was clearly shown to be the most promising.

Quality, Productivity and Economy in Welding Manufacturing - Case Study: West Africa

This thesis studies quality, productivity and economy in welding manufacturing in West African states such as Ghana, Nigeria and Cameroon. The study consists of two parts: the first part, which forms the theoretical background, reviews relevant literature concerning the metal and welding industries, and measurement of welding quality, productivity and economy. The second part, which is the empirical part, aims to identify activities in the metal manufacturing industries where welding is extensively used and to determine the extent of welding quality, productivity and economy measurements in companies operating in the metal manufacturing industries. Additionally, the thesis aims to identify challenges that companies face and to assess the feasibility of creating a network to address these issues. The research methods used in the empirical part are the case study (qualitative) method and the survey (quantitative) method. However, the case study method was used to elicit information from companies in Ghana, while the survey method was used to elicit information from companies in Nigeria and Cameroon. The study considers important areas that contribute to creating awareness and understanding of the current situation of the welding industry in West Africa. These areas include the metal manufacturing industrial sector, metal products manufactured, metal production and manufacturing systems deployed, welding quality, productivity and economy measurement systems utilized, equipment and materials on the markets, general challenges facing companies in welding operations, welding technology programs and research in local universities, and SWOT analysis of the various West African states. The notable findings indicate that majority of the companies operate in the constructionindustrial sector. Also, majority of the companies are project manufacturing oriented, thus provide services to customers operating in the growing industries such as the oil and gas, mining, food and the energy industry. In addition, only few companies are certified under standards such as ISO 9001, ISO 3834, and OHSAS 18001. More so, majority of the companies employ manual welding technique, and shielded metal arc welding (SMAW) as the commonly used welding process. Finally, welder salary is about € 300 / month as of June 2013 and the average operations turnover of medium to large companies is about € 5 million / year as at 2012. Based on analysis of the results of the study, it is noted that while welding activities are growing, the availability of cheap labor, the need for company and welder qualification and certification, and the need to manufacture innovative products through developmental projects (transfer of welding expertise and technology) remain as untapped opportunities in the welding industry in the West African states. The study serves as a solid platform for further research and concludes with several recommendations for development of the West African welding industry.

Lack of mutations of exon 2 of the MEN1 gene in endocrine and nonendocrine sporadic tumors

In addition to the mutations that underlie most cases of the multiple endocrine neoplasia type 1 (MEN1) syndrome, somatic mutations of the MEN1 gene have also been described in sporadic tumors like gastrinomas, insulinomas and bronchial carcinoid neoplasm. We examined exon 2 of this gene, where most of the mutations have been described, in 148 endocrine and nonendocrine sporadic tumors. DNA was obtained by phenol/chloroform extraction and ethanol precipitation from 92 formalin-fixed, paraffin-embedded samples, and from 40 fresh tumor tissue samples. We used 5 pairs of primers to encompass the complete coding sequence of exon 2 of the MEN1 gene that was screened by the polymerase chain reaction-single-stranded conformation polymorphism (PCR-SSCP) technique in 78 sporadic thyroid cancers: 28 follicular adenomas, 35 papillary carcinomas, 14 follicular carcinomas, and 1 anaplastic thyroid carcinoma. We also examined 46 adrenal lesions (3 hyperplasias, 3 adenomas and 35 adrenocortical carcinomas, 2 pheochromocytomas, 2 ganglioneuroblastomas, and 1 lymphoma) and 24 breast cancers (6 noninvasive, 16 infiltrating ductal, and 2 invasive lobular tumors). The PCR product of 5 tumors suspected to present band shifts by SSCP was cloned. Direct sense and antisense sequencing did not identify mutations. These results suggest that the MEN1 gene is not important in breast, thyroid or adrenal sporadic tumorigenesis. Because the frequency of mutations varies significantly among tumor subgroups and allelic deletions are frequently observed at 11q13 in thyroid and adrenal cancers, another tumor suppressor gene residing in this region is likely to be involved in the tumorigenesis of these neoplasms.

Antioxidant status of dog aqueous humor after extracapsular lens extraction

We determined the antioxidant status of the aqueous humor after extracapsular lens extraction in 14 mongrel dogs weighing about 10 kg. The animals were examined by slit lamp biomicroscopy, applanation tonometry and indirect ophthalmoscopy. One eye was submitted to conventional extracapsular lens extraction and the other was used as control. Samples of aqueous humor were obtained by anterior chamber paracentesis before and at days 1, 2, 3, 7 and 15 after surgery. Total antioxidant status was determined as the capacity of aqueous humor to inhibit free radical generation by 2,2-azobis(2-amidopropane) chlorine. Ascorbic acid concentration was measured by HPLC with UV detection. Protein content was determined with the biuret reagent. Statistical analysis was performed by ANOVA followed by the Tukey-Kramer test. Protein concentration increased from 0.61 to 22 mg/ml 24 h after surgery. These levels were maintained and returned to normal at day 7. Total antioxidant capacity was reduced from 50 to about 30 min until day 3 and at day 7 it was equal to control. Ascorbic acid levels were reduced from 252 to about 110 µM and then returned to control values at day 15. Considering the importance of ascorbic acid concentration in aqueous humor for the maintenance of the antioxidant status of the anterior segment of the eye, the decrease of antioxidant defenses suggests that the surgical procedures promote an oxidative stress condition in the eye.

Short-lasting systemic and regional benefits of early crystalloid infusion after intravenous inoculation of dogs with live Escherichia coli

We investigated the systemic and regional hemodynamic effects of early crystalloid infusion in an experimental model of septic shock induced by intravenous inoculation with live Escherichia coli. Anesthetized dogs received an intravenous infusion of 1.2 x 10(10) cfu/kg live E. coli in 30 min. After 30 min of observation, they were randomized to controls (no fluids; N = 7), or fluid resuscitation with lactated Ringer's solution, 16 ml/kg (N = 7) or 32 ml/kg (N = 7) over 30 min and followed for 120 min. Cardiac index, portal blood flow, mean arterial pressure, systemic and regional oxygen-derived variables, blood lactate, and gastric PCO2 were assessed. Rapid and progressive cardiovascular deterioration with reduction in cardiac output, mean arterial pressure and portal blood flow (~50, ~25 and ~70%, respectively) was induced by the live bacteria challenge. Systemic and regional territories showed significant increases in oxygen extraction and in lactate levels. Significant increases in venous-arterial (~9.6 mmHg), portal-arterial (~12.1 mmHg) and gastric mucosal-arterial (~18.4 mmHg) PCO2 gradients were also observed. Early fluid replacement, especially with 32 ml/kg volumes of crystalloids, promoted only partial and transient benefits such as increases of ~76% in cardiac index, of ~50% in portal vein blood flow and decreases in venous-arterial, portal-arterial, gastric mucosal-arterial PCO2 gradients (7.2 ± 1.0, 7.2 ± 1.3 and 9.7 ± 2.5 mmHg, respectively). The fluid infusion promoted only modest and transient benefits, unable to restore the systemic and regional perfusional and metabolic changes in this hypodynamic septic shock model.

Systemic and regional hemodynamic effects of enalaprilat infusion in experimental normotensive sepsis

Angiotensin-converting enzyme inhibitors have been shown to improve splanchnic perfusion in distinct shock states. We hypothesized that enalaprilat potentiates the benefits of early fluid resuscitation in severe experimental sepsis, particularly in the splanchnic region. Anesthetized and mechanically ventilated mongrel dogs received an intravenous infusion of live Escherichia coli over a period of 30 min. Thereafter, two interventions were performed: fluid infusion (normal saline, 32 mL/kg over 30 min) and enalaprilat infusion (0.02 mg kg-1 min-1 for 60 min) in randomized groups. The following groups were studied: controls (fluid infusion, N = 4), E1 (enalaprilat infusion followed by fluid infusion, N = 5) and E2 (fluid infusion followed by enalaprilat infusion, N = 5). All animals were observed for a 120 min after bacterial infusion. Mean arterial pressure, cardiac output (CO), portal vein blood flow (PVBF), systemic and regional oxygen-derived variables, and lactate levels were measured. Rapid and progressive reductions in CO and PVBF were induced by the infusion of live bacteria, while minor changes were observed in mean arterial pressure. Systemic and regional territories showed a significant increase in oxygen extraction and lactate levels. Widening venous-arterial and portal-arterial pCO2 gradients were also detected. Fluid replacement promoted transient benefits in CO and PVBF. Enalaprilat after fluid resuscitation did not affect systemic or regional hemodynamic variables. We conclude that in this model of normotensive sepsis inhibition of angiotensin-converting enzyme did not interfere with the course of systemic or regional hemodynamic and oxygen-derived variables.

Efficacy and safety of combined piroxicam, dexamethasone, orphenadrine, and cyanocobalamin treatment in mandibular molar surgery

Third molar extraction is a common procedure frequently accompanied by moderate or severe pain, and involves sufficient numbers of patients to make studies relatively easy to perform. The aim of the present study was to determine the efficacy and safety of the therapeutic combination of 10 mg piroxicam, 1 mg dexamethasone, 35 mg orphenadrine citrate, and 2.5 mg cyanocobalamin (Rheumazin®) when compared with 20 mg piroxicam alone (Feldene®) in mandibular third molar surgery. Eighty patients scheduled for removal of the third molar were included in this randomized and double-blind study. They received (vo) Rheumazin or Feldene 30 min after tooth extraction and once daily for 4 consecutive days. Pain was determined by a visual analogue scale and by the need for escape analgesia (paracetamol). Facial swelling was evaluated with a measuring tape and adverse effects and patient satisfaction were recorded. There was no statistically significant difference in facial swelling between Rheumazin and Feldene (control group). Both drugs were equally effective in the control of pain, with Rheumazin displaying less adverse effects than Feldene. Therefore, Rheumazin appears to provide a better risk/benefit ratio in the mandibular molar surgery. Since the side effects resulting from nonsteroidal anti-inflammatory drug administration are a severe limitation to the routine use of these drugs in clinical practice, our results suggest that Rheumazin can be a good choice for third molar removal treatment.

Corn germ with pericarp in relation to whole corn: nutrient contents, food and protein efficiency, and protein digestibility-corrected amino acid score

The germ fraction with pericarp (bran) is generated in the industrial processing of corn kernel, and it is used for oil extraction and animal feed. This study evaluated the nutritional and protein quality of this fraction in relation to whole corn. The proximate composition, mineral contents, and amino acid profile of the germ fraction with pericarp and of whole corn were determined. A 4-week experiment was conducted using 36 weanling male Wistar rats, and three 10%-protein diets (reference, germ with 15% lipids and casein with 15% lipids), two 6%-protein diets (whole corn and casein), and a protein-free diet were prepared. The germ showed higher contents of proteins, lipids, dietary fiber (27.8 g.100 g-1), ash, minerals (Fe and Zn- approximately 5 mg.100 g-1), and lysine (57.2 mg.g-1 protein) than those of corn. The germ presented good quality protein (Relative Protein Efficiency Ratio-RPER = 80%; Protein Digestibility-Corrected Amino Acid Score-PDCAAS = 86%), higher than that of corn (RPER = 49%; PDCAAS = 60%). The corn germ fraction with pericarp is rich in dietary fiber, and it is a source of good quality protein as well as of iron and zinc, and its use as nutritive raw material is indicated in food products for human consumption.

Heat Demand Forecasting Models’ Development: Use of Data Mining Tools in SQL Server Analysis Services

This thesis introduces heat demand forecasting models which are generated by using data mining algorithms. The forecast spans one full day and this forecast can be used in regulating heat consumption of buildings. For training the data mining models, two years of heat consumption data from a case building and weather measurement data from Finnish Meteorological Institute are used. The thesis utilizes Microsoft SQL Server Analysis Services data mining tools in generating the data mining models and CRISP-DM process framework to implement the research. Results show that the built models can predict heat demand at best with mean average percentage errors of 3.8% for 24-h profile and 5.9% for full day. A deployment model for integrating the generated data mining models into an existing building energy management system is also discussed.

The developmental state in Brazil: comparative and historical perspectives

The record of successful developmental states in East Asia and the partial successes of developmental states in Latin America suggest several common preconditions for effective state intervention including a Weberian bureaucracy, monitoring of implementation, reciprocity (subsidies in exchange for performance), and collaborative relations between government and business. Although Brazil failed to develop the high technology manufacturing industry and exports that have fueled sustained growth in East Asia, its developmental state had a number of important, and often neglected, successes, especially in steel, automobiles, mining, ethanol, and aircraft manufacturing. Where Brazil's developmental state was less successful was in promoting sectors like information technology and nuclear energy, as well as overall social and regional equality. In addition, some isolated initiatives by state governments were also effective in promoting particular local segments of industry and agriculture. Comparisons with East Asia, highlight the central role of state enterprises in Brazil that in effect internalized monitoring and reciprocity and bypassed collaboration between business and government (that was overall rarer in Brazil).

Data and Inventory Management in Spare Part Business: Developing Operations in the Case Company

After sales business is an effective way to create profit and increase customer satisfaction in manufacturing companies. Despite this, some special business characteristics that are linked to these functions, make it exceptionally challenging in its own way. This Master’s Thesis examines the current situation of the data and inventory management in the case company regarding possibilities and challenges related to the consolidation of current business operations. The research examines process steps, procedures, data requirements, data mining practices and data storage management of spare part sales process, whereas the part focusing on inventory management is reviewing the current stock value and examining current practices and operational principles. There are two global after sales units which supply spare parts and issues reviewed in this study are examined from both units’ perspective. The analysis is focused on the operations of that unit where functions would be centralized by default, if change decisions are carried out. It was discovered that both data and inventory management include clear shortcomings, which result from lack of internal instructions and established processes as well as lack of cooperation with other stakeholders related to product’s lifecycle. The main product of data management was a guideline for consolidating the functions, tailored for the company’s needs. Additionally, potentially scrapped spare part were listed and a proposal of inventory management instructions was drafted. If the suggested spare part materials will be scrapped, stock value will decrease 46 percent. A guideline which was reviewed and commented in this thesis was chosen as the basis of the inventory management instructions.

The development of automated methods for the determination of trace concentrations of carbomate pesticides in water using solid sorbent pre-concentration methods and high performance liquid chromatography /

Several automated reversed-phase HPLC methods have been developed to determine trace concentrations of carbamate pesticides (which are of concern in Ontario environmental samples) in water by utilizing two solid sorbent extraction techniques. One of the methods is known as on-line pre-concentration'. This technique involves passing 100 milliliters of sample water through a 3 cm pre-column, packed with 5 micron ODS sorbent, at flow rates varying from 5-10 mUmin. By the use of a valve apparatus, the HPLC system is then switched to a gradient mobile phase program consisting of acetonitrile and water. The analytes, Propoxur, Carbofuran, Carbaryl, Propham, Captan, Chloropropham, Barban, and Butylate, which are pre-concentrated on the pre-column, are eluted and separated on a 25 cm C-8 analytical column and determined by UV absorption at 220 nm. The total analytical time is 60 minutes, and the pre-column can be used repeatedly for the analysis of as many as thirty samples. The method is highly sensitive as 100 percent of the analytes present in the sample can be injected into the HPLC. No breakthrough of any of the analytes was observed and the minimum detectable concentrations range from 10 to 480 ng/L. The developed method is totally automated for the analysis of one sample. When the above mobile phase is modified with a buffer solution, Aminocarb, Benomyl, and its degradation product, MBC, can also be detected along with the above pesticides with baseline resolution for all of the analytes. The method can also be easily modified to determine Benomyl and MBC both as solute and as particulate matter. By using a commercially available solid phase extraction cartridge, in lieu of a pre-column, for the extraction and concentration of analytes, a completely automated method has been developed with the aid of the Waters Millilab Workstation. Sample water is loaded at 10 mL/min through a cartridge and the concentrated analytes are eluted from the sorbent with acetonitrile. The resulting eluate is blown-down under nitrogen, made up to volume with water, and injected into the HPLC. The total analytical time is 90 minutes. Fifty percent of the analytes present in the sample can be injected into the HPLC, and recoveries for the above eight pesticides ranged from 84 to 93 percent. The minimum detectable concentrations range from 20 to 960 ng/L. The developed method is totally automated for the analysis of up to thirty consecutive samples. The method has proven to be applicable to both purer water samples as well as untreated lake water samples.