No description available.

De nombreuses substances chimiques sont utilisées par l’industrie cosmétique pour entrer dans la composition de formules. En dehors de la nécessité d’évaluer leur efficacité, l’industrie cosmétique se doit surtout d’évaluer la sécurité de leurs substances pour l’humain. L'évaluation toxicologique des substances chimiques est réalisée dans le but de révéler un effet toxique potentiel de la substance testée. Parmi les effets potentiels que l’on souhaite détecter, la toxicité du développement (tératogénicité), c’est-à-dire la capacité d’une substance à provoquer l’apparition d’anomalies lors du développement embryonnaire, est fondamentale. En accord avec les législations internationales qui interdisent à l’industrie cosmétique d’avoir recours à des tests sur animaux de laboratoire pour l’évaluation de leurs substances, l’évaluation toxicologique de ces substances se base sur les résultats de tests in silico, in vitro et de tests faits sur des modèles alternatifs aux animaux de laboratoire. Pour le moment cependant, peu de méthodes alternatives existent et ont été validées pour la toxicologie du développement. Le développement de nouvelles méthodes alternatives est donc requis. D'autre part, en plus de l’évaluation de la sécurité des substances chez l’humain, l’évaluation de la toxicité pour l’environnement est nécessaire. L’usage de la plupart des produits cosmétiques et d’hygiène corporelle conduit, après lavage et rinçage, à un rejet à l’égout et donc dans les cours d’eau. Il en résulte que les environnements aquatiques (eaux de surface et milieux marins côtiers) sont parfois exposés aux substances chimiques incluses dans les formules cosmétiques. Ainsi, l’évaluation toxicologique environnementale des cosmétiques et de leurs ingrédients nécessite de connaître leur toxicité sur des organismes représentatifs de chaînes alimentaires aquatiques. Dans ce contexte, le modèle embryon de poisson présente un double avantage pour l’industrie cosmétique. Ce modèle, jugé par les législations internationales comme étant éthiquement acceptable pour les évaluations toxicologiques réalisées par l’industrie cosmétique, est représentatif des organismes aquatiques. Il est donc pertinent pour évaluer la toxicité environnementale des substances chimiques. D'autre part, ce modèle apparaît prometteur pour évaluer l’effet tératogène de substances chimiques chez l’humain. Pour ces raisons, un test d’analyse de la tératogénicité des substances chimiques est actuellement développé. Ce test se base sur l’analyse d’embryons de medaka (Oryzias Latipes) à 9 jours post fertilisation, après exposition des embryons par balnéation à des substances à concentrations déterminées. L’analyse de paramètres fonctionnels et morphologiques conduit au calcul d’un indice tératogène, qui permet de tirer une conclusion quant à l’effet tératogène de la substance testée. Cet indice est calculé à partir des mesures du taux de mortalité et du taux de malformations chez les embryons. L’objectif de ce projet est d’automatiser le test d’analyse de la tératogénicité, par classification automatique des embryons faite à partir d’image et de vidéo. La première méthode développée concerne la détection des battements cardiaques à partir de séquences vidéos, dans le but de calculer le taux de mortalité. Nous nous sommes ensuite concentrés sur deux types de malformations courantes qui sont les malformations axiales, et l'absence de vessie natatoire, en utilisant une méthode d'apprentissage automatique. Cette analyse doit être complétée par l'analyse d'autres malformations et conduire à un calcul du taux de malformations et de l’indice tératogène pour la substance testée / Numerous chemicals are used as ingredients by the cosmetics industry and are included in cosmetics formula. Aside from the assessment of their efficacy, the cosmetics industry especially needs to assess the safety of their chemicals for human. Toxicological screening of chemicals is performed with the aim of revealing the potential toxic effect of the tested chemical. Among the potential effects we want to detect, the developmental toxicity of the chemical (teratogenicity), meaning its capability of provoking abnormalities during the embryonic development, is crucial. With respect to the international regulations that forbid the use of animal testing for the safety assessment of cosmetics, the toxicological assessment of chemicals must base on an ensemble of in silico assays, in vitro assays and alternative models based assays. For now, a few alternative methods have been validated in the field of developmental toxicology. The development of new alternative methods is thus required. In addition to the safety assessment, the environmental toxicity assessment is also required. The use of most of cosmetics and personal care products leads to their rejection in waterways after washing and rince. This results in the exposition of some aquatic environments (surface waters and coastal marine environments) to chemicals included in cosmetics and personal care products. Thus, the environmental assessment of cosmetics and of their ingredients requires the knowledge of their toxicity on organisms that are representative of aquatic food chains. In this context, the fish embryo model, which is ethically acceptable according to international regulations, presents a dual advantage for the cosmetics industry. Firstly, as a model representative of aquatic organisms, it is accurate for the environmental assessment of chemicals. Secondly, this model is promising for the assessment of the teratogenic effect of chemicals on human. For this reason, a teratogenicity assessment test is developed. This test is based on the analysis of medaka fish embryos (Oryzias Latipes) at 9 days post fertilization, after balneation in a predetermined concentration of the chemical under study. The analysis of functional and morphological parameters allows to calculate a teratogenicity index, that depends on both rates of dead and malformed embryos. This index allows to to draw a conclusion concerning the teratogenic effect of the chemical.The objective of this project is to automate the teratogenicity test, by automated image and video classification. A first method is developed that aims to automatically detect embryo heart beats from acquired video sequences. This method will allow to calculate the proportion of dead embryos. We then focus on the detection of two common malformations: axial malformations and absence of a swim bladder, based on a machine learning classification. This analysis must be completed by the detection of other malformations so that we can measure the rate of malformed embryos and thus, calculate the teratogenicity index of the tested chemical

Although the descriptive representation of women in parliaments is continuously improving, sexist practices such as sexual harassment and domination techniques continue to permeate the inner workings of parliaments. No workplace is immune to sexual harassment but its prevalence in parliaments has serious implications, not only for those exposed but for democracy itself – conveying a message of who belongs in politics. While previous research has established the scope of the issue and its gendered and intersectional manifestations, little is known about parliamentary responses to sexual harassment. This study thus seeks to address this gap through a qualitative case study of the Swedish Parliament’s anti-harassment work, encompassing both Members of Parliament and parliamentary staff. By conducting qualitative text analysis on parliamentary documents and material gained through interviews with Swedish Members of Parliament and parliamentary staff, three dimensions of the Parliament’s anti-harassment work are explored. Initially, perceptions of sexual harassment are addressed, partly through the use of intersectional theory. Building on pioneer work in the field, de facto measures against sexual harassment are thereafter attended to. Lastly, by approaching parliaments as a specific form of gendered workplace with certain structural features, the difficulties inherent to the Swedish Parliament’s anti-harassment work are explored. Findings from the study indicate an awareness of sexual harassment as a gendered issue in the Parliament, albeit as a problem of limited scope and without recognition of how interactions between different social identity characteristics can further exposure. The mapping of different responses to sexual harassment through the three categories of regulations, complaint mechanisms and preventative/accompanying measures, reveals that measures are substantially more well developed for parliamentary staff and highlights that although there is active anti-harassment work in the Swedish Parliament, progressive efforts for Members of Parliament are continuously halted. While several difficulties are identified, the establishment of an independent complaint mechanism for Members of Parliament appears particularly pressing. Altogether, the findings indicate that the structural features of employment status, power and recurrent processes of socialising newcomers, integral to the parliamentary workplace, are important to consider when attempting to understand the disparities in measures between Members of Parliament and parliamentary staff and the difficulties in coming to terms with the issue.

Sleep is an important part of life as it affects the performance of one's activities during all awake hours. The study of sleep and wakefulness is therefore of great interest, particularly to the clinical and medical fields where sleep disorders are diagnosed. When studying sleep, it is common to talk about different types, or stages, of sleep. A common task in sleep research is to determine the sleep stage of the sleeping subject as a function of time. This process is known as sleep stage scoring. In this study, I seek to determine whether there is any benefit to using unsupervised feature learning in the context of electroencephalogram-based (EEG) sleep scoring. More specifically, the effect of generating and making use of new feature representations for hand-crafted features of sleep data – meta-features – is studied. For this purpose, two scoring algorithms have been implemented and compared. Both scoring algorithms involve segmentation of the EEG signal, feature extraction, feature selection and classification using a support vector machine (SVM). Unsupervised feature learning was implemented in the form of a dimensionality-reducing deep-belief network (DBN) which the feature space was processed through. Both scorers were shown to have a classification accuracy of about 76 %. The application of unsupervised feature learning did not affect the accuracy significantly. It is speculated that with a better choice of parameters for the DBN in a possible future work, the accuracy may improve significantly. / Sömnen är en viktig del av livet eftersom den påverkar ens prestation under alla vakna timmar. Forskning om sömn and vakenhet är därför av stort intresse, i synnerhet för de kliniska och medicinska områdena där sömnbesvär diagnostiseras. I forskning om sömn är det är vanligt att tala om olika typer av sömn, eller sömnstadium. En vanlig uppgift i sömnforskning är att avgöra sömnstadiet av den sovande exemplaret som en funktion av tiden. Den här processen kallas sömnmätning. I den här studien försöker jag avgöra om det finns någon fördel med att använda oövervakad inlärning av särdrag för att utföra elektroencephalogram-baserad (EEG) sömnmätning. Mer specifikt undersöker jag effekten av att generera och använda nya särdragsrepresentationer som härstammar från handgjorda särdrag av sömndata – meta-särdrag. Två sömnmätningsalgoritmer har implementerats och jämförts för det här syftet. Sömnmätningsalgoritmerna involverar segmentering av EEG-signalen, extraktion av särdragen, urval av särdrag och klassificering genom användning av en stödvektormaskin (SVM). Oövervakad inlärning av särdrag implementerades i form av ett dimensionskrympande djuptrosnätverk (DBN) som användes för att bearbetasärdragsrymden. Båda sömnmätarna visades ha en klassificeringsprecision av omkring 76 %. Användningen av oövervakad inlärning av särdrag hade ingen signifikant inverkan på precisionen. Det spekuleras att precisionen skulle kunna höjas med ett mer lämpligt val av parametrar för djuptrosnätverket.

This work presents a scalable solution to automate game-testing. Traditionally, game-testing has been performed by either human players or scripted Artificial Intelligence (AI) agents. While the first produces the most reliable results, the process of organizing testing sessions is time consuming. On the other hand, scripted AI dramatically speeds up the process, however, the insights it provides are far less useful: these agents’ behaviors are highly predictable. The presented solution takes the best of both worlds: the automation of scripted AI, and the richness of human testing by framing the problem within the Deep Reinforcement Learning (DRL) paradigm. Reinforcement Learning (RL) agents are trained to adapt to any unseen level and present customizable human personality traits: such as aggressiveness, greed, fear, etc. This is achieved exploring the problem from a multi-task RL setting. Each personality trait is understood as a different task which can be linearly combined by the proposed algorithm. Furthermore, since Artificial Neural Networks (ANNs) have been used to model the agent’s policies, the solution is highly adaptable and scalable. This thesis reviews the state of the art in both automatic game-testing and RL, and proposes a solution to the above-mentioned problem. Finally, promising results are obtained evaluating the solution on two different environments: a simple environment used to quantify the quality of the designed algorithm, and a generic game environment useful to show-case its applicability. In particular, results show that the designed agent is able to perform good on game levels never seen before. In addition, the agent can display any convex combination of the trained behaviors. Furthermore, its performance is as good as if it had been specifically trained on that particular combination. / Detta arbete presenterar en skalbar lösning för att automatisera speltestning. Traditionellt har speltestning utförts av antingen mänskliga spelare eller förprogrammerade agenter. Även om det förstanämnda ger de mest tillförlitliga resultaten är processen tidskrävande. Å andra sidan påskyndar förprogrammerade agenter processen dramatiskt, men de insikter som de ger är mycket mindre användbara: dessa agenters beteenden är mycket förutsägbara. Den presenterade lösningen använder det bästa av två världar: automatiseringsmöjligheten från förprogrammerade agenter samt möjligheten att simulera djupet av mänskliga tester genom att inrama problemet inom paradigmet Djup Förstärkningsinlärning. En agent baserad på förstärkningsinlärning tränas i att anpassa sig till tidigare osedda spelmiljöer och presenterar anpassningsbara mänskliga personlighetsdrag: som aggressivitet, girighet, rädsla... Eftersom Artificiella Neurala Nätverk (ANNs) har använts för att modellera agentens policyer är lösningen potentiellt mycket anpassnings- och skalbar. Denna rapport granskar först den senaste forskningen inom både automatisk speltestning och förstärkningsinlärning. Senare presenteras en lösning för ovan nämnda problem. Slutligen evalueras lösningen i två olika miljöer med lovande resultat. Den första miljön används för att kvantifiera kvaliteten på den designade algoritmen. Den andra är en generisk spelmiljö som är användbar för att påvisa lösningens tillämplighet.

Prostate cancer is the most common cancer in Sweden. Patients with the condition have a good prognosis in general and most cases can be treated. Localized prostate cancer is primarily treated via surgery or radiation therapy and is diagnosed with the help of different imaging modalities, such as magnetic resonance imaging, MRI, and positron emission tomography, PET. The diagnosis is confirmed and the aggressiveness of the cancer is determined through biopsies. Samples from a small part of the prostate are extracted and then examined. This could mean that parts of higher aggressiveness may be missed, which in turn could lead to under-treatment of the cancer. The aggressiveness of a lesion can be described by Gleason Score, GS, which is determined by an visual assessment of the shape, size and arrangement of the cells. The aim of this study was to correlate GS with in-vivo images using MRI and PET. This was accomplished by investigating image data from PSMA PET, Acetate PET, Ktrans MRI and T2-weighted MRI from a cohort of 26 prostate cancer patients containing 74 lesions. Regions of interests, ROI:s, were created and applied on all images. Statistics such as median and max value were extracted from each ROI. The statistics were combined to get a wide range of descriptive variables for each respective imaging modality. These were normalised against a certain zone of the prostate or only the absolute value. The results indicated that PSMA PET, Acetate PET and Ktrans MRI were correlated to GS, while T2-weighted MRI was not. Data also indicated that PSMA PET, Acetate PET and Ktrans MRI give complementary information to each other, which could indicate that a combination of the modalities would better predict GS. The implications of these findings could affect both the diagnostics and the treatment of prostate cancer.

Syftet med denna studie är att undersöka hur undervisning med genrepedagogiska drag påverkar elevers argumenterande texter i åk 6. Studien utgår från tre frågeställningar som ligger till grund för undersökningen. Dessa frågor rör vilka genredrag elever använder i argumenterande texter före undervisning, hur undervisningen genomförs och vilka genredrag elever använder i argumenterande texter efter undervisningen. Materialet för studien baseras på observationer av genrepedagogiskt baserad undervisning och elevtextanalyser av argumenterande texter skrivna av elever i årskurs 6. Till grund för studien ligger den sociokulturella teorin där läraren stöttar eleverna i sitt lärande samt cykeln för undervisning och lärande. För att analysera texterna före och efter undervisning används en checklista med genretypiska drag för den argumenterande genren. Studien visar tydligt att före undervisning med genrepedagogiska drag har eleverna svårigheter med att skriva väl fungerande argumenterande texter. Med stöd av undervisning om genredrag i argumenterande texter utvecklar eleverna sitt skrivande och skriver i högre grad texter med genrespecifika drag. Dock visar studien att ytterligare stöttning av läraren och mer tid för att lära sig de genretypiska dragen för en text behövs för att eleverna ska klara av att göra egna kreativa val i sina texter.

This master thesis studies the learning of dense feature descriptors where camera poses are the only supervisory signal. The use of camera poses as a supervisory signal has only been published once before, and this thesis expands on this previous work by utilizing a couple of different techniques meant increase the robustness of the method, which is particularly important when not having access to ground-truth correspondences. Firstly, an adaptive robust loss is utilized to better differentiate inliers and outliers. Secondly, statistical properties during training are both enforced and adapted to, in an attempt to alleviate problems with uncertainties introduced by not having true correspondences available. These additions are shown to slightly increase performance, and also highlights some key ideas related to prediction certainty and robustness when working with camera poses as a supervisory signal. Finally, possible directions for future work are discussed.

A novel approach for a robust computation of positional relations of two-dimensional geometric features is presented which guarantees reliable results, provided that the initial data is valid. The method is based on the use of integer coordinates and a method to generate a complete, gap-less and non-overlapping spatial decomposition. The spatial relationships of two geometric features are then represented using DE-9IM matrices. These allow the spatial relationships to be represented compactly. The DE-9IM matrices are based on the spatial decomposition using explicit neighborhood relations. No further geometric calculations are required for their computation. Based on comparative tests, it could be proven that this approach, up to a predictable limit, provides correct results and thus offers advantages over classical methods for the calculation of spatial relationships. This novel method can be used in all fields, especially where guaranteed reliable results are required.:Introduction Related Research Materials and Methods Results Discussion Outlook Author Contributions Funding Institutional Review Board Statement Informed Consent Statement Data Availability Statement Conflicts of Interest Abbreviations References

E-commerce has seen a rapid growth the last two decades, making it easy for customers to shop wherever they are. The growth has also led to new kinds of fraudulent activities affecting the customers. To make customers feel safe while shopping online, companies like Resurs Bank are implementing different kinds of fraud filters to freeze transactions that are thought to be fraudulent. The latest type of fraud filter is based on machine learning. While this seems to be a promising technology, data and algorithms need to be tuned properly to the task at hand. This thesis project gives a proof of concept of realizing a machine learning based fraud filter for Resurs Bank. Based on a literature study, available data and explainability requirements, this work opts for a supervised learning approach based on Random Forests with a sliding window to overcome concept drift. The inherent class imbalance of the setting makes the area-under-the-receiver operating-curve a suitable metric. This approach provided promising results that a machine learning based fraud filter can add value to companies like Resurs Bank. An alternative approach on how to incorporate non-numerical features by using recurrent neural networks (RNN) was implemented and compared. The non-numerical feature was transformed by a pre-trained RNN-model to a numerical representation that reflects the features suspiciousness. This new numerical feature was then included in the Random Forest model and the result demonstrated that this approach can add valuable insight to the fraud detection field.