För små företag utan rekryteringsansvarig person kan det var svårt att hitta rätt personal. Brist på sådana resurser är en påfrestning som leder till stress och mindre lyckade rekryteringar. Målet med arbetet var att hitta en lösning för att automatisera matchning i en rekryteringsprocess genom att ge förslag på relevanta personer som tidigare sökt jobb hos företag via Cheffle:s tjänst. Det finns flera olika sätt att matcha uppsättningar av data. I det här fallet användes maskininlärning som lösningsmetod. Detta implementerades tillsammans med en prototyp som hämtade in data om jobbet och den arbetssökande. Maskininlärningsmodellerna Supportvektormaskin och Artificiella Neural Nätverk använde sig av denna data för att betygsätta de arbetssökande. Detta utifrån hur väl de matchade jobbannonsen. Arbetets slutsats är att ingen modell hade tillräckligt hög precision i sina klassificeringar för att användas i en verklig implementation, detta då endast små mängder testdata fanns tillgänglig. Resultatet visade att maskininlärningsmodellerna Supportvektormaskin och Artificiella Neurala Nätverk visade potential för att kunna användas vid denna typ av matchning, men för att fastställa detta krävs mer träningsdata / It can be hard for a small company with no dedicated HR-role to find suiting recruits. A lack of resources takes a toll on the existing employees and increase stress that further harms recruiting. The goal of this work was to find a solution to automate matching in a recruitment process by suggesting relevant applicants that have previously used Cheffle. There are multiple ways of matching data. In the case of this study, machine learning was used. A prototype was developed. It collected data about a job and its related applicants. The data was then used by the machine learning models Support vector machine and Artificial Neural Network to classify the applicants by how closely they match the job position. The conclusion made in this work is that no model had a precision high enough in its classification to be used in a final implementation. The low precision in classification is likely a result of the small amount of test data available. The result showed that the machine learning models Support vector machine and Artificial Neural Network had potential in this type of matching. To firmly determine this the models would need to be tested with more test data.

Regression models were developed to predict scenic quality for residential streets in Ann Arbor, Michigan for both Summer and Winter vegetative conditions. Scenic quality was quantified using the Scenic Beauty Estimation method. Only variables that existed in the city's computer data base were used. Variables such as diameter at breast height, basal area, number of trees, and tree species diversity were investigated as to their predictive ability. In addition, the predictive ability of quadratic, power, inverse, and logarithmic transformations of these variables was investigated. The best predictive Summer model used the natural log of the average diameter of street trees and the natural log of the average assessed property value as variables. The best predictive Winter model used the natural log of the average diameter of street trees as its independent variable. / Master of Science / incomplete_metadata

Diferentes abordagens teóricas têm sido utilizadas em estudos de sistemas biomoleculares com o objetivo de contribuir com o tratamento de diversas doenças. Para a dor neuropática, por exemplo, o estudo de compostos que interagem com o receptor sigma-1 (Sig-1R) pode elucidar os principais fatores associados à atividade biológica dos mesmos. Nesse propósito, estudos de Relações Quantitativas Estrutura-Atividade (QSAR) utilizando os métodos de regressão por Mínimos Quadrados Parciais (PLS) e Rede Neural Artificial (ANN) foram aplicados a 64 antagonistas do Sig-1R pertencentes à classe de 1-arilpirazóis. Modelos PLS e ANN foram utilizados com o objetivo de descrever comportamentos lineares e não lineares, respectivamente, entre um conjunto de descritores e a atividade biológica dos compostos selecionados. O modelo PLS foi obtido com 51 compostos no conjunto treinamento e 13 compostos no conjunto teste (r² = 0,768, q² = 0,684 e r²teste = 0,785). Testes de leave-N-out, randomização da atividade biológica e detecção de outliers confirmaram a robustez e estabilidade dos modelos e mostraram que os mesmos não foram obtidos por correlações ao acaso. Modelos também foram gerados a partir da Rede Neural Artificial Perceptron de Multicamadas (MLP-ANN), sendo que a arquitetura 6-12-1, treinada com as funções de transferência tansig-tansig, apresentou a melhor resposta para a predição da atividade biológica dos compostos (r²treinamento = 0,891, r²validação = 0,852 e r²teste = 0,793). Outra abordagem foi utilizada para simular o ambiente de membranas sinápticas utilizando bicamadas lipídicas compostas por POPC, DOPE, POPS e colesterol. Os estudos de dinâmica molecular desenvolvidos mostraram que altas concentrações de colesterol induzem redução da área por lipídeo e difusão lateral e aumento na espessura da membrana e nos valores de parâmetro de ordem causados pelo ordenamento das cadeias acil dos fosfolipídeos. As bicamadas lipídicas obtidas podem ser usadas para simular interações entre lipídeos e pequenas moléculas ou proteínas contribuindo para as pesquisas associadas a doenças como Alzheimer e Parkinson. As abordagens usadas nessa tese são essenciais para o desenvolvimento de novas pesquisas em Química Medicinal Computacional. / Different theoretical approaches have been used in the studies of biomolecular systems aiming to contribute with the treatment of several diseases. For neuropathic pain, for example, the study of compounds that interact with sigma-1 receptor (Sig-1R) can elucidate the main factors associated to their biological activities. For this purpose, studies of Quantitative Structure-Activity Relationships (QSAR) using Partial Least Squares (PLS) and Artificial Neural Network (ANN) methods were applied to 64 Sig-1R antagonists belong to 1-arylpyrazole class. PLS and ANN models were used in order to describe linear and nonlinear behavior, respectively, between a set of descriptors and the biological activity of the selected compounds. The PLS model was obtained with 51 compounds in the training set and 13 compounds in the test set (r² = 0.768, q² = 0.684 and r²test = 0.785). Leave-N-out tests, biological activity randomization and outliers detection confirmed the robustness and stability of the models and showed that they were not obtained by chance correlations. Models were also generated from Multilayer Perceptron Artificial Neural Network (MLP-ANN) and the 6-12-1 architecture, trained by tansig-tansig transfer functions, showed the best result for the biological activity prediction of the compounds (r²training = 0.891, r²validation = 0.852 and r²test = 0.793). Another approach was used to simulate synaptic membranes environment using lipid bilayers composed by POPC, DOPE, POPS and cholesterol. Performed molecular dynamics studies showed that high cholesterol concentration induces decrease of area per lipid and lateral diffusion and increase of membrane thickness and order parameter caused by ordering of phospholipids acil chains. The obtained lipid bilayers can be used to simulate interactions between lipids and small molecules or proteins contributing for researches associated to Alzheimer and Parkinson diseases. The approaches used in this thesis are essential for the development of new researches in Computational Medicinal Chemistry.

Presidentvalet i USA baseras på personval istället för partival och detta personval inkluderar även presidentkandidatens partner. Presidentkandidaternas fruar diskuteras och analyseras idag lika mycket som presidentkandidaterna själva samtidigt som medierna diskuterar vilken av de tidigare presidentfruarna de kommer att efterlikna. Syftet med den här uppsatsen är att undersöka och analysera vilken retorik Michelle Obama och Ann Romney använder sig av i talen de ger på The Democratic National Convention respektive The Republican National Convention i ett försök att övertyga de amerikanska väljarna att deras man borde bli deras nästa president. Eftersom det visuella framträdandet bidrar till talets helhet ska även det undersökas och analyseras. Då Michelle Obama och Ann Romney har jämförts under hela valet görs det även här en komparativ analys av deras retorik samt deras visuella framträdande. En kvalitativ metod har valts för uppsatsen i form av en retorisk analys. Denna retoriska analys görs med hjälp av Karlbergs och Mrals (1998) retoriska modell. Resultatet visar att både Michelle Obama och Ann Romney använder sig mest av ethos och pathos i sina tal för att bygga tillit till publiken samt visa känslor. Undersökningens resultat visar även att de båda blir presidentkandidaternas surrogat till målgrupper som de inte når ut till vilket visar att presidentskapet är en två- persons- karriär där presidentfrun får stå för den feminina sidan medan presidentposten förbli maskulin. Resultatet visar även att den politiska kommunikationen inte behöver vara strikt politisk och att medierna spelar en stor roll i den politiska kommunikationen idag. / The presidential election in The United States is based on voting for an individual candidates rather than voting for a party and this vote includes the wife of the presidential candidate. The wives of presidential candidates are today equally discussed and analyzed, while the media also discusses which one of the previous presidential wives they are going to take after. The purpose of this thesis is to examine and analyze the rhetoric that Michelle Obama and Ann Romney uses in the speeches they give at The Democratic National Convention and The Republican National Convention in an effort to convince the American voters that their husband should be their next president. Since the visual appearance contributes to the speech as a whole this will also be examined and analyzed. Michelle Obama and Ann Romney have been compared throughout the whole election so a comparative analysis of their rhetoric and their visual appearance will also be made here. A qualitative approach has been chosen for this thesis in the form of a rhetorical analysis. This rhetorical analysis is done by using Karlberg and Mrals (1998) rhetorical model.. The results show that both Michelle Obama and Ann Romney used ethos and pathos the most in their speeches to build trust with the audience and to show emotions. The thesis results also show that they are surrogates to the presidential candidates in the way that they can reach target groups that the candidates cannot and with this showing that the presidency is a two-person career where the presidential wife to stands for the feminine side while the presidential post remains masculine. The results also show that political communication doesn’t have to be purely political, and that the media plays a major role in political communication today.

Fundação de Amparo a Pesquisa no Estado do Rio Grande do Sul / In the late eighteenth century, Ann Radcliffe established herself as one of the most famous novelists of her time, and she reached the peak of her career with her fourth novel entitled The Mysteries of Udolpho (1794). Although it is one of the most important gothic novels, this narrative has issues not much explored by critics yet. The Mysteries of Udolpho presents us with one of the strongest characteristics of Radcliffe s fiction, namely, the detailed construction of setting. In this sense, this work analyses and discusses the roles of setting, which is organized in the novel as natural or constructed setting. The analysis focuses on the relation between this thematic-formal aspect and the actions and personal relationships of the protagonist Emily with other characters. The discussion shows that the different types of setting are essential in the narrative once they have strong participation in the ideological construction of characters regarding the connection between sentimentalism and rationality. Therefore, the relation between setting and characters, in this particular novel, expresses important aspects of the complex development of the bourgeois sensibility in eighteenth-century England. / Em fins do século XVIII, Ann Radcliffe se estabeleceu como uma das romancistas mais famosas de sua época, atingindo o ápice de sua carreira com seu quarto romance intitulado The Mysteries of Udolpho (1794). Apesar de ser um dos romances góticos ingleses mais importantes, ele ainda apresenta questões pouco exploradas pelos críticos. The Mysteries of Udolpho possui uma das características mais fortes das obras de Radcliffe: a minuciosa elaboração do espaço. Em vista disso, este estudo analisa e discute as funções do espaço, o qual está organizado em natural e construído. A análise centra na maneira como esse aspecto temático-estrutural se relaciona com as ações e relações pessoais da protagonista Emily com as demais personagens. Discutimos como os diferentes tipos de espaço tornam-se essenciais por participarem de forma enfática na construção ideológica das personagens, no que diz respeito à associação entre sentimentalismo e racionalidade. Assim, a relação entre espaço e personagens nesse romance expressa aspectos importantes da complexa construção da sensibilidade burguesa na Inglaterra do século XVIII.

Diferentes abordagens teóricas têm sido utilizadas em estudos de sistemas biomoleculares com o objetivo de contribuir com o tratamento de diversas doenças. Para a dor neuropática, por exemplo, o estudo de compostos que interagem com o receptor sigma-1 (Sig-1R) pode elucidar os principais fatores associados à atividade biológica dos mesmos. Nesse propósito, estudos de Relações Quantitativas Estrutura-Atividade (QSAR) utilizando os métodos de regressão por Mínimos Quadrados Parciais (PLS) e Rede Neural Artificial (ANN) foram aplicados a 64 antagonistas do Sig-1R pertencentes à classe de 1-arilpirazóis. Modelos PLS e ANN foram utilizados com o objetivo de descrever comportamentos lineares e não lineares, respectivamente, entre um conjunto de descritores e a atividade biológica dos compostos selecionados. O modelo PLS foi obtido com 51 compostos no conjunto treinamento e 13 compostos no conjunto teste (r² = 0,768, q² = 0,684 e r²teste = 0,785). Testes de leave-N-out, randomização da atividade biológica e detecção de outliers confirmaram a robustez e estabilidade dos modelos e mostraram que os mesmos não foram obtidos por correlações ao acaso. Modelos também foram gerados a partir da Rede Neural Artificial Perceptron de Multicamadas (MLP-ANN), sendo que a arquitetura 6-12-1, treinada com as funções de transferência tansig-tansig, apresentou a melhor resposta para a predição da atividade biológica dos compostos (r²treinamento = 0,891, r²validação = 0,852 e r²teste = 0,793). Outra abordagem foi utilizada para simular o ambiente de membranas sinápticas utilizando bicamadas lipídicas compostas por POPC, DOPE, POPS e colesterol. Os estudos de dinâmica molecular desenvolvidos mostraram que altas concentrações de colesterol induzem redução da área por lipídeo e difusão lateral e aumento na espessura da membrana e nos valores de parâmetro de ordem causados pelo ordenamento das cadeias acil dos fosfolipídeos. As bicamadas lipídicas obtidas podem ser usadas para simular interações entre lipídeos e pequenas moléculas ou proteínas contribuindo para as pesquisas associadas a doenças como Alzheimer e Parkinson. As abordagens usadas nessa tese são essenciais para o desenvolvimento de novas pesquisas em Química Medicinal Computacional. / Different theoretical approaches have been used in the studies of biomolecular systems aiming to contribute with the treatment of several diseases. For neuropathic pain, for example, the study of compounds that interact with sigma-1 receptor (Sig-1R) can elucidate the main factors associated to their biological activities. For this purpose, studies of Quantitative Structure-Activity Relationships (QSAR) using Partial Least Squares (PLS) and Artificial Neural Network (ANN) methods were applied to 64 Sig-1R antagonists belong to 1-arylpyrazole class. PLS and ANN models were used in order to describe linear and nonlinear behavior, respectively, between a set of descriptors and the biological activity of the selected compounds. The PLS model was obtained with 51 compounds in the training set and 13 compounds in the test set (r² = 0.768, q² = 0.684 and r²test = 0.785). Leave-N-out tests, biological activity randomization and outliers detection confirmed the robustness and stability of the models and showed that they were not obtained by chance correlations. Models were also generated from Multilayer Perceptron Artificial Neural Network (MLP-ANN) and the 6-12-1 architecture, trained by tansig-tansig transfer functions, showed the best result for the biological activity prediction of the compounds (r²training = 0.891, r²validation = 0.852 and r²test = 0.793). Another approach was used to simulate synaptic membranes environment using lipid bilayers composed by POPC, DOPE, POPS and cholesterol. Performed molecular dynamics studies showed that high cholesterol concentration induces decrease of area per lipid and lateral diffusion and increase of membrane thickness and order parameter caused by ordering of phospholipids acil chains. The obtained lipid bilayers can be used to simulate interactions between lipids and small molecules or proteins contributing for researches associated to Alzheimer and Parkinson diseases. The approaches used in this thesis are essential for the development of new researches in Computational Medicinal Chemistry.

Generellt spelande är ett forskningsområde fokuserat på att skapa AI som kan spela spel utan någon domänspecifik information. Detta arbete har undersökt elman-nätverks potential för generellt Atari-spelande genom att testa ett elman-nätverk och ett feedforward-nätverk via the Arcade Learning Environment. Nätverken använder en pixelrepresentation för att representera spelmiljön och baserar sina handlingar endast på den informationen. Agenterna testades på fyra spel varav två anses kräva en mer avancerad struktur än feedforward. Agenterna evalueras via deras toppoäng i spelen som testas och tränas via en genetisk algoritm. Resultaten visade att elman-strukturen inte presterar bättre än feedforward, dessutom erhölls ingen poäng i de avancerade spelen vilket tyder på att ett korttidsminne inte är tillräckligt för att spela dessa spel. Jämfört med tidigare forskning sågs en liten förbättring över liknande struktur vilket tyder på en förbättrad representation. För att förbättra resultaten i detta arbete borde ett större antal spel testas.

Block partitioning is a computationally heavy step in the video coding process. Previously, this stage has been done using a full-search-esque algorithm. Recently, Artificial Neural Networks (ANN) approaches to speed-up block partitioning in encoders compliant to the Versatile Video Coding (VVC) standard have shown to significantly decrease the time needed for block partitioning. In this degree project, a state of the art Convolutional Neural Network (CNN) was ported to VTM16. It was ablated into 7 new models which were trained and tested. The eects of the ablations were compared and discussed with respect to the number of Multiply-Accumulate operations (MAC) a model required, the speed-up in the encoding stage as well as the quality of the encoding. The results show that the number of MACs can be substantially decreased from that of the state of the art model while having low negative eects on the quality of the encoding. Furthermore, the results show that the two tested approaches of reducing the computational complexity of the model were eective. Those were: 1) reducing the image’s resolution earlier in the model. 2) reducing the number of features in the beginning layers. The results point towards the first approach being more eective. / Blockpartitionering är ett beräkningstungt steg i videokodningsprocessen. Tidigare har detta gjorts genom att använda en algoritm i fullsökningsstil. Nyligen har artificiella neurala nätverk (ANN) visats vara eektiva för att minska tidsåtgången för blockpartitioneringen i enkodare som följer Versatile Video Coding-standarden (VVC). I detta examensarbete har en framgångsrik Convolutional Neural Networkmodell (CNN) portats till VTM16. Stegvisa ändringar på denna modell har gjorts för att ta fram sju modeller som tränades och testades. Eekten av ändringarna på ursprungsmodellen jämfördes och diskuterades med hänsyn till antalet Multiply-Accumulate-operationer (MAC) som respektive modell krävde, deras påverkan på tidsåtgången samt deras påverkan på kvalitén av kodningen. Resultaten visar att antalet MACs kan minskas betydligt utan att betydelsefullt minska kvalitén på kodningen. Resultaten visar att de båda testade tillvägagångssätt för att minska beräkningskomplexiteten var eektiva. Tillvägagångssätten var 1) minska bildens upplösning i ett tidigare skede i modellen. 2) minska antalet kanaler i de tidigare lagren. Resultaten pekar mot att det första tillvägagångssättet är mer eektivt.

On the 29th of August, the vessel Stena Scandica experienced a blackout. Before the blackout, 294 alarms were issued in 4 minutes. With the number of alarms, the operators could not prevent the blackout. The amount of information and the way it was presented became a hindrance to operators. They could not interpret their surroundings' information without fault from them. This interpretation is called situational awareness. This thesis will solve how information can be provided to operators without hindrance to situational awareness. The focus will be on the Swedish Navy's operators and their needs. The aim is to solve the problem by creating a system that provides situational awareness. The system will use the information on air- and seaborne targets from a radar and a camera display. Three research questions were proposed: how will the radar data structure be, how will it be ranked, and how will it be presented? The structure was expected to tell the targets' location, size, and movement. The ranking of the targets would tell if the targets were a threat to the naval operators. Lastly, the targets were expected to be presented with some of their information on a camera display.    For the first question, the structure for both kinds of targets was constructed to meet the expectations. Two models were used to solve the second question. An artificial neural network and fuzzy c-means. The artificial neural network was chosen as it is one of the best classification algorithms. Fuzzy c-means were chosen since it can cluster similar behaviors together, therefore clustering high-threat targets together. Of these two models, the result showed that the artificial neural network was a better ranking method, with a higher accuracy of 92.9% for airborne targets and 80.6% for seaborne targets. A simulation was made to answer the third question and was built according to the expectations. The simulation only displayed the highest threat targets in the camera display. By presenting the high-threat targets, the operators received a better understanding of where the targets are in reality. In the future, studies should be conducted on implementation of the system on Swedish Navy vessels. For example, is there enough computational power for an artificial neural network? / Den 29 augusti drabbades fartyget Stena Scandica av ett strömavbrott. Innan strömavbrottet utlöste 294 larm inom 4 minuter, vilket gjorde det omöjligt för operatörerna att förhindra avbrottet. Mängden information och sättet den presenterades på blev ett hinder för operatörerna, vilket påverkade deras lägesbild. Arbete syftar till att lösa hur information kan tillhandahållas till operatörer utan att hindra deras situationsmedvetenhet, med fokus på den svenska marinens operatörer och deras behov. Detta arbete föreslår ett system som använder radardata och kameradisplayer för att tillhandahålla lägesbilden. Tre forskningsfrågor ställs: hur ska radarns datastruktur vara, hur ska den rankas, och hur ska den presenteras? Strukturen förväntas visa målens plats, storlek och rörelse. Rankningen ska indikera om målen utgör ett hot, och hög-hotmål ska presenteras på kameradisplayen. För att svara på den första frågan konstruerades strukturen för båda typerna av mål. För den andra frågan användes två modeller: ett artificiellt neuralt nätverk och fuzzy c-means. Det artificiella neurala nätverket visade sig vara den bästa metoden med en noggrannhet på 92,9% för luftmål och 80,6% för sjömål. En simulering gjordes för att svara på den tredje frågan, där endast de mest hotfulla målen visades på kameradisplayen. Detta gav operatörerna en bättre förståelse för var målen befann sig. Framtida studier bör undersöka systemets implementering på svenska marinens fartyg. Exempelvis om tillräcklig beräkningskraft finns för ett artificiellt neuralt nätverk.

Master of Science / Department of Civil Engineering / Yacoub M. Najjar / Reliable and economical design of Portland Cement Concrete (PCC) pavement structural systems relies on various factors, among which is the proper characterization of the expected permeability response of the concrete mixes. Permeability is a highly important factor which strongly relates the durability of concrete structures and pavement systems to changing environmental conditions. One of the most common environmental attacks which cause the deterioration of concrete structures is the corrosion of reinforcing steel due to chloride penetration. On an annual basis, corrosion-related structural repairs typically cost millions of dollars. This durability problem has gotten widespread interest in recent years due to its incidence rate and the associated high repair costs. For this reason, material characterization is one of the best methods to reduce repair costs. To properly characterize the permeability response of PCC pavement structure, the Kansas Department of Transportation (KDOT) generally runs the Rapid Chloride Permeability test to determine the resistance of concrete to penetration of chloride ions as well as the Boil test to determine the percent voids in hardened concrete. Rapid Chloride test typically measures the number of coulombs passing through a concrete sample over a period of six hours at a concrete age of 7, 28, and 56 days. Boil Test measures the volume of permeable pore space of the concrete sample over a period of five hours at a concrete age of 7, 28, and 56 days. In this research, backpropagation Artificial Neural Network (ANN)-based and Regression-based permeability response prediction models for Rapid Chloride and Boil tests are developed by using the databases provided by KDOT in order to reduce or eliminate the duration of the testing period. Moreover, another set of ANN- and Regression-based permeability prediction models, based on mix-design parameters, are developed using datasets obtained from the literature. The backpropagation ANN learning technique proved to be an efficient methodology to produce a relatively accurate permeability response prediction models. Comparison of the prediction accuracy of the developed ANN models and regression models proved that ANN models have outperformed their counterpart regression-based models. Overall, it can be inferred that the developed ANN-Based permeability prediction models are effective and applicable in characterizing the permeability response of concrete mixes used in transportation applications.