<p>This thesis aims to design and develop the spatial adaptation approach through spatial transformers to improve the accuracy of human keypoint recognition models. We have studied different model types and design choices to gain an accuracy increase over models without spatial transformers and analyzed how spatial transformers increase the accuracy of predictions. A neural network called Widenet has been leveraged as a specialized network for providing the parameters for the spatial transformer. Further, we have evaluated methods to reduce the model parameters, as well as the strategy to enhance the learning supervision for further improving the performance of the model. Our experiments and results have shown that the proposed deep learning framework can effectively detect the human key points, compared with the baseline methods. Also, we have reduced the model size without significantly impacting the performance, and the enhanced supervision has improved the performance. This study is expected to greatly advance the deep learning of human key points and movement dynamics. </p>

This paper uses a back-propagating neural network (BPN) to predict the price movements of major crypto currencies, leveraging technical factors as well as measurements of collective sentiment derived from the micro-blogging network Twitter. Our dataset consists of daily, hourly and minutely price levels for Bitcoin, Ether and Litecoin along with 8 popular technical indicators, as well as all tweets with the currencies' cash tags during respective time periods. Insprired by previous research which suggest that artificial neural networks are superior forecasting models in this setting, we were able to create a system generating automated investment decisions on a daily, hourly and minutely time basis. The study concluded that price trends are indeed predictable, with a correct prediction rate above 50% for all models, and corrensponding profitable trading strategies for all currencies on an hourly basis when neglecting trading fees, buy-sell spreads and order delays. The overall highest predictability is obtained on the hourly trading interval for Bitcoin, yielding an accuracy of 55.74% and a cumulative return of 175.1% between October 16, 2021 and December 31, 2021. / I denna studie används ett bakåtpropagerande neoronnät (BPN) för att förutsäga prisrörelser i större kryptovalutor med hjälp av tekniska faktorer och kvantifiering av kollektivt sentimentet från mikrobloggnätverket Twitter. Vårt dataset består av dagliga, timvisa och minutvisa prisnivåer för Bitcoin, Ether och Litecoin tillsammans med 8 populära tekniska indikatorer, samt alla tweets med valutornas "cash tags" under respektive tidsperiod. Med inspiration från tidigare forskning som hävdar att artificiella nauronnät är överlägsna prognosmodeller i denna typ av analys kunde vi skapa ett system som genererar automatiska investeringsbeslut på daglig, timvis och minutvis basis. Vi hävdar med denna studie att pristrender är förutsägbara för dessa kryptovalutor, med en korrekt förutsägelsefrekvens på över 50% för alla modeller, och med lönsamma handelsstrategier för alla valutor på timbasis när man bortser från handelsavgifter, köp- och säljspreadar och orderfördröjningar. Den högsta förutsägbarheten erhålls på timhandelsintervallet för Bitcoin, vilket ger en nogrannhet på 55,74% och en ackumulerad avkastning på 175,1% mellan den 16 oktober 2021 och den 31 december 2021.

<p dir="ltr">Measuring human kinematics and kinetics is critical for ergonomists to evaluate ergonomic risks related to physical workloads, which are essential for ensuring workplace health and safety. Human kinematics describes human body postures and movements in 6 degrees of freedom (DOF). In contrast, kinetics describes the external forces acting on the human body, such as the weight of loads being handled. Measuring them in the workplace has remained costly as they require expensive equipment, such as motion capture systems, or are only possible to measure manually, such as measuring the weight through a force gauge. Due to the limitations of existing measurement methods, most ergonomics assessments are conducted in laboratory settings, mainly to evaluate and improve the design of workspaces, production tools, and tasks. Continuous monitoring of workers' ergonomic risks during daily operations has been challenging, yet it is critical for ergonomists to make timely decisions to prevent workplace injuries.</p><p dir="ltr">Motivated by this gap, this dissertation proposed three studies that introduce novel low-cost, minimally intrusive, and automated methods to measure human kinematics and kinetics for ergonomics assessments. Specifically, study 1 proposed ErgoNet, a deep learning and computer vision network that takes a monocular image as input and predicts the absolute 3D human body joint positions and rotations in the camera coordinate system. It achieved a Mean Per Joint Position Error of 10.69 cm and a Mean Per Joint Rotation Error of 13.67 degrees. This study demonstrated the ability to measure 6 DOF joint kinematics for continuous and dynamic ergonomics assessments for biomechanical modeling using just a single camera. </p><p dir="ltr">Studies 2 and 3 showed the potential of using pressure-sensing gloves (i.e., tactile gloves) to predict ergonomics risks in lifting tasks, especially the weight of loads. Study 2 investigated the impacts of different lifting risk factors on the tactile gloves' pressure measurements, demonstrating that the measured pressure significantly correlates with the weight of loads through linear regression analyses. In addition, the lifting height, direction, and hand type were found to significantly impact the measured pressure. However, the results also illustrated that a linear regression model might not be the best solution for using the tactile gloves' data to predict the weight of loads, as the weight of loads could only explain 58 \% of the variance of the measured pressured, according to the R-squared value. Therefore, study 3 proposed using deep learning model techniques, specifically the Convolution Neural Networks, to predict the weight of loads in lifting tasks based on the raw tactile gloves' measurements. The best model in study 3 achieved a mean absolute error of 1.58 kg, representing the most accurate solution for predicting the weight of loads in lifting tasks. </p><p dir="ltr">Overall, the proposed studies introduced novel solutions to measure human kinematics and kinetics. These can significantly reduce the costs needed to conduct ergonomics assessments and assist ergonomists in continuously monitoring or evaluating workers' ergonomics risks in daily operations.</p>

Considérant un point de vue général de cette thèse aborde le problème de trouver, à partir d'un ensemble de blocs de construction, un sous-ensemble qui procure une solution à un problème donné. Ceci est fait en tenant compte de la compatibilité de chacun des blocs de construction par rapport au problème et l'aptitude d'interaction entre ces parties pour former ensemble une solution. Dans la perspective notamment de la thèse sont les blocs de construction de méta-modèles et le problème donné est une description d'un problème peut être résolu en utilisant un logiciel et d'être résolu en utilisant un système multi-agents. Le noyau de la proposition de thèse est un processus qui analyse un problème donné et puis il proposé une solution possible basée sur système multi-agents pour ce problème. Il peut également indiquer que le problème ne peut être résolu par ce paradigme. Le processus adressée par la thèse consiste en les étapes principales suivantes: (1) A travers un processus de caractérisation on analyse la description du problème pour localiser le domaine de solutions, puis choisissez une liste de candidats des méta-modèles. (2) Les caractérisations de méta-modèles candidats sont prises, ils sont définis dans plusieurs domaines de la solution. On fait la chois parmi le domaine trouvé dans la étape précédant. (3) On crée un système multi-agents où chaque agent représente un candidat méta-modèle. Dans cette société les agents interagissent les uns avec les autres pour trouver un groupe de méta-modèles qui est adapté pour représenter une solution donnée. Les agents utilisent des critères appropriés pour chaque méta-modèle à représenter. Il évalue également la compatibilité des groupes créés pour résoudre le problème de décider le groupe final qui est la meilleure solution. Cette thèse se concentre sur la fourniture d'un processus et un outil prototype pour résoudre plutôt la dernière étape de la liste. Par conséquent, le chemin proposé a été créé à l'aide de plusieurs concepts de la méta-analyse, l'intelligence artificielle de coopération, de la cognition bayésienne, incertitude, la probabilité et statistique. / From a general point of view this thesis addresses an automatic path to build a solution choosing a compatible set of building blocks to provide such a solution to solve a given problem. To create the solution it is considered the compatibility of each available building block with the problem and also the compatibility between each building block to be employed within a solution all together. In the particular perspective of this thesis the building blocks are meta-models and the given problem is a description of a problem that can be solved using software using a multi-agent system paradigm. The core of the thesis proposal is the creation of a process based on a multi-agent system itself. Such a process analyzes the given problem and the available meta-models then it matches both and thus it suggests one possible solution (based on meta-models) for the problem. Nevertheless if no solution is found it also indicates that the problem can not be solved through this paradigm using the available meta-models. The process addressed by the thesis consists of the following main steps: (1) Through a process of characterization the problem description is analyzed in order to locate the solution domain and therefore employ it to choose a list of most domain compatible meta-models as candidates. (2) There are required also meta-model characterization that evaluate each meta-model performance within each considered domain of solution. (3) The matching step is built over a multi-agent system where each agent represents a candidate meta-model. Within this multi-agent system each agent interact with each other in order to find a group of suitable meta-models to represent a solution. Each agent use as criteria the compatibility between their represented candidate meta-model with the other represented meta-models. When a group is found the overall compatibility with the given problem is evaluated. Finally each agent has a solution group. Then these groups are compared between them in order to find the most suitable to solve the problem and then to decide the final group. This thesis focuses on providing a process and a prototype tool to solve the last step. Therefore the proposed path has been created using several concepts from meta-analysis, cooperative artificial intelligence, Bayesian cognition, uncertainty, probability and statistics.

長久以來,人類在萬物中獨具的高智慧特質吸引了無數的哲學家和科學家 投入對其研究,除了醫學的原因之外,由於人腦所具有卓越的辨識系統及學 習能力,為數不少的科學家們相信人腦存在許多最適化系統與設計,因此如 何模仿人類腦神經的組織與運作,一直是很多人努力及夢寐以求的.因此類 神經網路就是依據這些理念而在各研究領域上廣為發展與應用,其中本文 所探討的倒傳遞神經網路模型更是目前類神經網路模型中最具代表性,應 用最廣的模型.而結構性時間數列模型則是將可被觀察的變數分解成趨勢, 季節性,不規則性等不可被觀察項,故其對經濟意義的解釋是相當明當明顯 的,藉由狀態空間模式的轉換,我們將很容易地利用卡門濾器來作估計與預 測.而本文所欲探的重點在於比較有學習機能的倒傳遞神經網及可利用最 新的資訊更新之結構性時間數列何者之預測能利較佳,藉此瞭解二者之一 些特性.

Syfte Syftet med studien är att undersöka fotbollsutövande gymnasieelevers uppfattningar av skaderisk på konstgräs respektive naturgräs. Metod En kvantitativ enkätundersökning med 46 stycken (st) deltagande gymnasieelever som läser lokal (LIU) eller nationell (NIU) godkänd idrottsutbildning inom fotboll. Wilcoxon Signed Rank Test användes för beräkning av medelvärde och för att urskilja statistiska signifikanta skillnader mellan de olika grupperna, match och träning på respektive spelunderlag. Chi- Square crosstabs post-hoc test användes för att urskilja statistiska signifikanta skillnader mellan de olika skadetyperna på respektive spelunderlag. Resultat Resultaten i denna studie visade att de deltagande fotbollsutövande gymnasieeleverna uppfattade att skaderisken för akuta- överbelastningsskador var högre på konstgräs än naturgräs, både under träning och match. Skillnaden mellan de olika grupperna, akuta- och överbelastningsskador under träning och match på respektive spelunderlag, var statistiskt signifikanta då signifikans nivån var mindre än p&lt;0,05. De skadetyperna som de deltagande fotbollsutövande gymnasieeleverna uppfattade vara mest förekommande på konstgräs var rivsår/ skrapsår. De skadetyperna som de deltagande fotbollsutövande gymnasieeleverna uppfattade vara mest förekommande på naturgräs var muskelsträckning/ muskelbristning. Skillnaden mellan de olika grupperna, muskelbristning/ muskelsträckning samt rivsår/ skrapsår på konstgräs och naturgräs, var statistiskt signifikanta då signifikans nivån var mindre än p&lt;0,05. Slutsatser Denna studie bekräftar att de deltagande fotbollsutövande gymnasieelevernas uppfattningar stämmer överens med vad tidigare studier har visat om elitfotbollsspelares uppfattningar om skaderisk på konstgräs i jämförelse med naturgräs. Att skaderisken för akuta- och överbelastningsskador uppfattas öka vid spel på konstgräs i jämförelse med naturgräs. Kvalitativa forskningsmetoder skulle kunna bidra till djupare förståelse av själva ämnet, samt upplevelser av uppfattningar av skaderisk och skadetyper på respektive spelunderlag. För att fler slutsatser ska kunna dras bör framtida studier även inkludera fler deltagare. / Purpose The aim with this study is investigate football practicing high school students’ perceptions of risk of injury on artificial turf and natural grass. Methods A quantitative questionnaire survey including 46 participating football practicing high school students. The Wilcoxon Signed Rank Test was used to calculate the mean and to distinguish statistically significant differences between the different groups, match and training on artificial turf and natural grass. The Chi-Square crosstab post-hoc test was used to distinguish statistically significant differences between the different types of injury on the respective game grounds Results The results in this study showed that the participating football practicing high school students’ perceived that the injury risk of acute- and overload injuries was higher on artificial turf than on natural grass, both during training and match. The difference between the different groups, acute and overload injuries during training and match on articial turf and natural grass was statistically significant as the level of significancewas less than p &lt;0.05. The injury types that the participating football practicing high school students perceived to be the most common on artificial turf was abrasion/ laceration. The injury types that the participating football practicing high school students perceived to be the most common on natural grass were musclesprain/ strain. The difference between the different groups, musclesprain/ strain as well as abrasion/ laceration on artificial turf and natural grass was statistically significant as the level of significance was less than p &lt;0.05. Conclusions This study confirms that the participating football practicing high school students’ perceptions concur with what previous studies have shown about elite football players' perceptions of injury risk on artificial turf in comparison with natural grass. The risk of acute and overload injuries is perceived to increase when playing on artificial turf in comparison to natural grass. Qualitative research methods could contribute to deeper understanding of the subject itself, as well as experiences of perceptions of injuries and types of injury on respective turfs. In order to draw further conclusions, should future studies include more participants.

The graduation thesis "Analysis of stock-exchange data using AI methods" is focused on the use of neural networks while predicating the exchange-rate movements on Change. The theoretical part is divided into three independent units. The Change matters and the related individual terms are described in the first part. In the second part, the two basic approaches to the stock-exchange data analysis are analyzed, these two approaches being the fundamental and technical analysis. The third, and the last, theoretical part forms an individual unit describing the Artificial Intelligence theory. Particularly the issue of the neuronal networks is described in detail. The practical part seeks the use for the chosen neuronal network GAME. It analyses the chosen YMZ9 market. It focuses on the prediction of the exchange-rate movements using the "sliding window" method. The last chapter summarizes the results and it proves that under certain circumstances it is possible to properly use the neuronal networks both for the prediction of the stock-exchange movements and as one of the corner-stones of the profitable trading system.

This diploma thesis deals with image super-resolution with conservation of good quality. Firstly, there are described state of the art methods dealing with this problem, as well as principles of neural networks with focus on convolutional ones. Finally, there is described a few models of convolutional neural network for image super-resolution to double size, which have been trained, tested and compared on newly created database with pictures of people.

This master thesis deals with classification of sleep stages on the base of polysomnographic signals. On several signals was performed analysis and feature extraxtion in time domain and in frequency domain as well. For feature extraxtion was used EEG, EOG and EMG signals. For classification was selected classification models K-NN, SVM and artifical neural network. Accuracy of classifation is different depending on used method and spleep stages split. The best results achieved classification among stages Wake, REM, and N3, with neural network usage. In this case the succes was 93,1 %.

The diploma thesis deals with artificial intelligence utilization for predictions on stock markets.The prediction is unconventionally based on Bayes' probabilistic model theorem and on its based Naive Bayes classifier. I the practical part algorithm is designed. The algorithm uses recognized relations between identifiers of technical analyze. Concretely exponential running averages at 20 and 50 days had been used. The program output is a graphic forecast of future stock development which is designed on ground of relations classification between the identifiers