In the area of Author Identication many approaches have been made to identify the author of a written text. By identifying the individual variation that can be found in texts, features can be calculated. These feature values are commonly calculated by normalizing the values to an average valueover the whole text. When using this kind of Simple features much of the variation that can be found in texts will not get captured. This project intends to use the sequential nature of the text to denie Sequential featuresat sentence level. The theory is that the Sequential features will be able to capture more of the variation that can be found in the texts, compared to the Simple features. To evaluate these features a classication of authors was made on several dierent datasets. The result showed that the Sequential features performs better than the Simple features in some cases, however the dierence was not large enough to confirm the theory of them being better than the Simple features. / Inom området som behandlar författarbestämning har många olika tillvägagångs- sätt använts for att identiera författaren av en skriven text. Genom att identfiera den individuella variation som särskiljer texter från varandra,kan olika särdrag beräknas. Dessa särdrags värden beräknas vanligen genom att normaliseras till ett medelvärde över hela texten. När denna typ av Enkla särdrag används så döljs mycket av den variation som särskiljertexter från varandra. Målet med detta projekt är att istället användatextens sekventiella natur som grund for att deniera Sekventiella särdrag på meningsnivå. Teorin är att de sekventiella särdragen kommer att kunna identifiera mer av den variation som kan identifieras i texter, jämfört med de enkla särdragen. For att utvärdera dessa särdrag gjordes en klassicering av författare på era olika dataset. Resultatet visade att de sekventiella särdragen presterade bättre än de enkla särdragen i vissa fall, men skillnaden var inte tillräckligt stor for att bekräfta teorin om att de skulle vara bättre än de enkla sädragen.

This thesis explores hand pose estimation, which means mapping a 2D image to a hand pose. Hand pose estimation has many promising applications such as hand sign recognition, robotic learning by demonstration, and human-computer-interaction in general. To do the estimation, image features are extracted from the image and a mapping to the space of hand poses is then constructed. Ideally the mapping from image features to pose space would be one-to-one, but in reality it is rather a many-to-many mapping leading to ambiguities. This is due to the image feature not capturing the actualpose, but the form of the 2D projection of the hand. Hands may also occludeparts of itself which also leads to ambiguities. This thesis explores waysin which to improve the commonly used image feature HOG (Histogram of Oriented Gradients), by capturing the HOG subspace used by hand images,to obtain a feature whose mapping to pose space is more well-behaved than that of the HOG feature. The new feature is computed as projection on and distances to lines in HOG space. The new feature’s performance is tested against the HOG feature using nearest neighbour (NN) regression and the results show that the new feature does not yet perform as well as the HOG feature. Never the less, the conclusion is that the new feature, called MPDD, for Multiple Projection and Distance Dimensions, does indeed capture the most relevant information in HOG, but fail to use it as well as the HOG does with the current construction method. However, constructing the MPDD ina slightly different way could potentially lead to improvements and so future research could still be of interest. / Denna masteruppsats undersöker handposestimering där en 2D bild används för att rekonstruera en handpos, vilken beskrivs av hur handlederna är vinklade. Handposestimering har många potentiella användningsområden varav några är teckenspråksigenkänning, robotinlärning från demonstrationer, men även människa-datorinteraktion i allmänhet. För att göra översättningen från bild till handpos extraheras först bildegenskaper (image-features) varpå en mappning till posrummet (alla möjliga poser) konstrueras. Den önskade egenskapen hos en mappning till posrummet är framförallt att den är one-to-one,men i verkligheten innehåller den normalt tvetydigheter. Detta beror bland annat på att bild- egenskaperna inte beskriver själva handposen utan enbart formen av 2D projektionen av en hand. Förutom det så kan händer även skymma delar av sig själva vilket också leder till tvetydigheter i mappningen från bildegenskaper till posrummet. Målet med denna uppsats är ändå att undersöka sätt att förbättra den vanligt använd bildegenskap HOG (Histogramof Oriented Gradients) genom att använda det delrum hos HOG deskriptorerna som upptas av handbilder. Den nya bildegenskapen konstrueras genom projektioner och avståndsberäkningar till linjer i HOG rummet som motsvarar handrörelser. Den föreslagna bildegenskapen testas mot HOG med NN(Nearest neighbour) regression och resultatet visar att HOG presterar bäst med den nuvarande konstruktionen av den nya bildegenskapen. Slutsatsen är dock att den nya deskriptorn, kallad MPDD för Multiple Projection and Distance dimensions, lyckas fånga det relevanta delrummet av HOG, men misslyckas med att använda informationen i denna. Detta innebär sannolikt att sättet som MPDD konstrueras på i denna mastersuppsats antagligen bör förändras även om en liknande idé kan räcka för att uppnå samma eller bättre resultat än HOG.

Recommender systems have been widely adopted by onlinee-commerce websites like Amazon and music streaming services like Spotify. However, most research efforts have not sufficiently considered the context in which recommendations are made, especially when the input is implicit. In this work, we investigate the value of including contextual information like day-of-week in collaborative filtering recommender systems. For the investigation, we first implemented two algorithms, namely contextual prefiltering and contextual post-filtering. Then, we evaluated these algorithms with user data collected from Spotify. Experiment results show that the pre-filtering algorithm shows some promise against an item similarity baseline, indicating that further investigation could be rewarding. The post-filtering algorithm underperforms a popularity-derived baseline, due to information loss in the recommendationprocess.

Pervasive games are games played in the real world with various support from mobile and sensor technology. It is becoming much simpler to commercially stage pervasive games due to the high level of penetration of smart mobile phones. However, technology supported pervasive games are seldom fully automated games. There are two main reasons for this phenomenon, many of the engaging game activities such as the physical encounter and the interaction between players cannot be fully sensed and captured by technology. Secondly, the physical world where in pervasive games are staged provides an ever-changing scene, a scene that interacts and interferes with pre-programmed functionalities. For these reasons, many pervasive games require game-mastering functions.The goal of this project is to study the interfaces and systems to provide the game-mastering functions in the context of pervasive games using the game Codename: Heroes as its use case. As a result we present a generic Service Oriented Architecture for game-mastering interfaces of pervasive games, and a semi-generic interface that uses a visualization tool that combines the spatial, temporal and social domains. The visualization tool explores the manipulation of time-ranges in a timeline to facilitate the exploration of the time domain.

Categorising bank transactions to predened categories are essential for getting a good overview of ones personal nance. Tink provides a mobile app for automatic categorisation of bank transactions. Tink's categorisation approach is a clustering technique with longest prex match based on merchant.This thesis will examine if a machine learning model can learn to classify transactions based on its purchase, what was bought, instead of merchant.This thesis classies bank transactions in a supervised learning setting by exploring early and late fusion schemes on three types of modalities (text, amount,date) found in Swedish bank transactions. Experiments are carried out with Naive Bayes, Support Vector Machines and Decision Trees. The dierent fusionschemes are compared with no fusion, learned on only one modality, and stacked classication, learning models in a pipe-lined fashion.The early fusion concatenation schemes shows all worse performance than no fusion on the text modality. The late fusion experiments on the other hand shows no impact of modality fusion.Suggestions are made to change the feedback loop from user, to get more data labeled by users, which would potentially boost the other modalities importance / Att sköta sin privatekonomi med hjälp av kategorisering gör nog många människor omedvetet, en försöker helt enkelt få en känsla pa vad en lägger sina pengar på. För att kunna ge full översikt på hur ens privatekonomi ser ut, har Tink skapat en mobilapplikation for att automatiskt kategorisera banktransaktioner. Detta görs just nu med klustering och längsta prex matchning på forsäljningsställe. Kategoriseringen av banktransaktioner ger användaren en direkt återkoppling på hur pengaflödet ser ut samt till vad och när dessa köp görs. Den har uppsatsen kommer att undersoka om en maskininlärningsmodell kan lära sig att klassicera banktransaktioner baserat pa köp istället för försäljningsställe. Genom att undersöka två olika fusioneringsscheman på tre typer av modaliteter funna i banktransaktioner (text, pris och datum), ska vi forsoka uttröna dessa modaliteters påverkan på klassicering. De olika scheman är jamförda med ingen fusionering, dvs inlärning på endast en modalitet, och travad klassicering,dvs inlärning med era efterföljande modeller.Experimenten ar gjorda med supervised-learning och inlärningsmodellerna är Naive Bayes, Support Vector Machines samt Beslutstrad. Experimenten visar på att klassicering på text, alltså försäljningsställe ger bäst resultat i jämförelse med alla de andra experimenten. I de tidiga fusionsexperimenten visar alla modalitet-sammanslagningar sämre resultat än ingen fusion på bara text. De sena fusion experimenten visar å andra sidan ingen skillnad alls efter fusionering med modaliteternas pris och datum. Förslag på förbättrad klassicering på köp antas öka, alltså modaliteternas pris och datum bör vara mer betydande, om mer var datamärkt av användare.

Eye tracking is a field that has been growing immensely over the last decade. Accompanying this growth is a need for simplified and automatic analysis of eye tracking data. A part of that analysis is eye movement classification, and while there are many adequate classification methods for fixations and saccades, the tools for smooth pursuit classification are still lacking. This thesis gives an overview of the field,and analyses five different methods for classifying smooth pursuits, fixations,and saccades. The analysis also explores evaluation methods that avoid the laborious way of manually tagging data to get a reference classification. Despite earlier reports of decent performance, the overall results for all the analysed algorithms is poor. In particular, the slowest pursuits are consistently misclassified. Most certainly, the inclusion of the slow pursuits have skewed the results, but even disregarding them doesn’t yield particularly impressive results. This begs the question of what concessions one has to make in terms of prerequisites on the data, or qualifiers for the resulting analysis, to achieve adequate performance,and given those, when would such a classification be preferred to something tailored to the problem at hand?

Den här rapporten ämnar undersöka vilket det huvudsakliga särdraget hos musikalisk genre är.En avgränsning har gjorts till hypotesen att enbart melodi är tillräckligt för att klassificera genre alltså tt melodi är det huvudsakliga särdraget hos musikalisk genre. En kvantitativ metod har använts i form av en internetbaserad tjänst där användare har fått klassificera genre på musikstycken i sin helhet, samma musikstycken med extraherad sång samt samma musikstycken representerade i syntetiserade versioner vilket representerar melodierna för musikstyckena. Statistiska metoder har sedan använts för att presentera den insamlade datan,för att efter analys komma fram till att melodi ej är tillräckligt för att klassificera genre och alltsåatt melodi ej är det huvudsakliga särdraget hos musikalisk genre. / This thesis aims to investigate the main deterministic feature of musical genre. A delimitation has been made to the hypothesis that solely melody is sufficient to determine musical genre; i.e.that melodi is the main deterministic feature of musical genre. A qualitative method has beenused in form of an internet based service where users were asked to determine genre on wholemusical piece, the same musical pieces with the lead song extracted as well as the samemusical pieces represented as synthesized versions whichrepresents the melodies of themusical pieces. Statistical methods were used in order to present the gathered data, and after data analysis a conclusion was conducted that melody is not sufficient in order to determine genre, which means that melody is not the main deterministic feature of musical genre.

A listening experiment was conducted to investigate which musical instruments are the most important for defining certain musical genres. 66 participants genre classified a series of audio samples, with the same songs recurring both with full instrumentation and partial instrumentation. The report used the collected genre classifications to clarify therelationship between certain musical genres and song instrumentation. A numericalanalysis of the classifications, in the context of genre traditions and conventions, show that certain traditions hold true, while others do not. The most and least defining instrumentation for each genre was determined and discussed / Ett lyssningsexperiment genomfördes för att undersöka vilka musikinstrument som är de mest centrala för att definiera en särskild musikgenre. 66 testpersoner fick klassificera ett antal ljudexpempel efter genre. Samma låtar återkom med både full och delvis instrumentering. Rapporten använde de resulterande genreklassificeringarna för attförtydliga sambandet mellan musikgenrerna och instrumentering. En numerisk analys avresultaten utfördes och analyserades i ett sammanhang av olika musikgenre traditioner. Det visade sig att vissa traditioner överenssstämmer med den numeriska analysen, medan andra traditioner inte gör det. Den mest och minst genre- definerande stämman beräknades och sammanställdes i en tabell.

The opportunity of vitalizing recourses is becoming increasingly attractive, and traditional machine virtualization is no longer the only thing in mind. Today, virtualization also concerns resources such as firewalls, routers and switches. For this to be possible and successful, a flexible and evolutionary scheduler is necessary. The focus of this thesis project has been to explore and develop a generic cross-service scheduler between the computing service Nova and networking service Neutron. We have designed a generic foundation for a scheduler and the aim in the future is to use this solution for other services in OpenStack such as volume, storage and image. Our starting point was the filter scheduler in Nova, which originally is designed for scheduling virtual machines (VMs) rather than scheduling network resources. A unique cross-service scheduler between Nova and Neutron is implemented and discussed in this paper. This thesis presents the design, implementation and functional verification of the Neutron scheduler and cross-service communication. The cross-service scheduling is designed so that Nova and Neutron are loosely coupled. By loosely coupled we mean that Nova and Neutron are as independent of each other as possible. The report concludes with a discussion about the project process and implementation process, and the how the solution can be developed further e.g. by implementing CLI commands for increasing the functionality of filter chains.

The aim of the work is to design and prototype an autonomous stroke rehabilitation system using the Microsoft Kinect camera that allows patients to undergo their rehab therapy from home without the constant need of specialized staff involvement in the rehab process. The rehab system tracks and computes a set of measurable indicators that reflect the rehab progress. Today, patients that have suffered a stroke usually have to visit specialized centers to perform the rehabilitation program. This proves to be hard for the patients with motion disorders, especially in the northern parts of Sweden with large distances to the closest specialized center. A prototype rehabilitation system for stroke patients has been designed and built. The system is autonomous and does not need constant staff involvement in the rehab process. The system tracks a set of rehabilitation indicators that reflect the patient rehabilitation progress (joint range of motion, reaction time, precision of motion, energy expenditure and training time). The system is constantly monitoring the patient to make sure the exercises are done correctly. Attention has been paid at making the system more engaging and fun by adding some gamification features like providing real time feedback while exercising and by computing the training statistics with personal best indicators and progress meant to increase patient engagement and motivation. At last but not at least, the system has a multimodal interface including audio feedback that makes usage much more intuitive and simple. The system was designed and implemented and tested on regular users. The results prove that the system is able to achieve good results in automating the rehabilitation process and providing