För att underlätta grafikers animationsarbete har det länge varit intressant att utforska olika fysikbaserade lösningar och göra dem effektiva. Detta arbetes mål har varit att undersöka olika villkorslösare för trasdocksfysik och fastställa vilken som hade bäst balans mellan tidseffektivitet och trovärdighet. Bakgrunden presenterar tidigare forskning som arbetet bygger på och går igenom viktiga begrepp inom trasdocksfysiksimulering relaterat till funktionalitet och trovärdighet. Olika scenarier och en benchmarker utvecklades för att testa trasdocksversionerna och kortare videoklipp skapades som observerades av respondenter för att avgöra versionernas trovärdighet. Resultatet från undersökningen visade inte på någon större skillnad för trovärdigheten av versionerna, men visade en större skillnad av tidseffektivitet mellan dem. Den sammantagna bedömningen blev att versionen med tre relaxationssteg, utan prioriteringsprocess av förskjutningar hade bäst balans. För framtida arbete är det intressant att studera hur väl versionerna presterar om de körs på GPU och att lägga till självkollision för trasdockan. / <p>Det finns övrigt digitalt material (t.ex. film-, bild- eller ljudfiler) eller modeller/artefakter tillhörande examensarbetet som ska skickas till arkivet.</p>

Background. Onion services facilitate two-way communication over the Tor net-work without letting either party know the other address or location. Many different techniques to break that anonymizing have come forth, but most of them have only been on paper. Some have been tested but then only on a separate network and not on the live Tor network. Objectives. This thesis presents a technique that, with a minimal intrusion to the Tor network and no manipulation of the Introduction relay or the Rendezvous relay,can break the anonymizing of an Onion service. Methods. The technique has been tested on the live Tor network with the approval of an ethics board. The Onion service anonymity was broken by having the Guard relay the Onion service used to connect to the Tor network introducea watermark containing the IP4 address of the Onion service in the TCP packet’s Request-Response Time (RRT). The TCP packets were used to transmit the water-mark where an HTTP echo request was sent from a Tor client where the RRT was captured, and the watermark was decoded. In order to decode the watermark, the normal RRT of packets on the Tor network was needed, so to get the data, HTTP echo requests were also sent without the watermark. Results. The watermark was decoded by the Tor client 88.80% of the time out of 607 tries. Conclusions. While this technique was proven to work, what holds it back is the need for the Onion service to choose the Guard relay that introduces the watermark.The chance of a specific Guard relay is chosen depends on that relays history on the Tor network. However, it’s usually about 0.005%, meaning it would need around,20000 tries to break the anonymity of a random Onion service if only one Guard relay is used.

In tropical and subtropical countries with bad infrastructure there exists diseases which are often neglected and untreated. Some of these diseases are caused by parasitic intestinal worms which most often affect children severely. The worms spread through parasite eggs in human stool that end up in arable soil and drinking water. Over one billion people are infected with these worms, but medication is available. The problem is the ineffective diagnostic method hindering the medication to be distributed effectively. In the process of designing an automated microscope for increased effectiveness the solution for marking out the stool sample on the microscope slide is important for decreasing the time of diagnosis. This study examined the active contour model and four different semantic segmentation networks for the purpose of delineating the stool sample from the other parts of the microscope slide. The Intersection-over-Union (IoU) measurement was used to measure the performance of the models. Both active contour and the networks increased the IoU compared to the current implementation. The best model was the FCN-32 network which is a fully convolutional network created for semantic segmentation tasks. This network had an IoU of 95.2%, a large increase compared to the current method which received an IoU of 77%. The FCN-32 network showed great potential of decreasing the scanning time while still keeping precision of the diagnosis.

Modern tools in the forest management industry lack the features to properly visualizeforests. This study explores the possibilities of recreating real-world forests in a three-dimensional virtual environment to give users the opportunity to remotely view whattheir forest looks like. To keep accessibility high, public map data was used. In this studya proof-of-concept application was developed in Unity using Mapbox’s SDK. Theaccuracy of the virtual world was then evaluated by comparing it to photographs fromthe real world and a point cloud render. Furthermore, performance was evaluated bymeasuring load times and frame rates when the application ran. The results show that itis possible to create a forest that resembles the real one when viewed from afar, butdetails such as individual tree positions and tree species were inaccurate. High-endcomputers with dedicated graphics card ran the application efficiently, while low-endhardware struggled.

Dementia is a syndrome of illnesses resulting in cognitive decline, severely impacting the lives of those afflicted as well as their loved ones. The most common form of dementia is Alzheimer's disease, with roughly 10 million new cases each year. In this study we examine different machine learning models and approaches aimed to aid healthcare professionals in early diagnosis of Alzheimer's disease, potentially automating parts of the diagnostic process. We evaluate our models on the Pitt corpus of the Dementia Bank dataset, using 10-fold cross validation. We compare the BERT and RoBERTa transformer models, and find that both models achieve high accuracy, precision, and specificity. The highest accuracy is achieved by RoBERTa, reaching an accuracy of 86.72%, a precision of 90.69% and a specificity of 90.53%. Furthermore, we explore the viability of using automated speech recognition for automatic transcription of audio samples from patient meetings. RoBERTa achieves an accuracy of 83.59% using transcripts generated by Google's automatic speech recognition, suggesting such methods may be viable for automating certain parts of the diagnostic process. In addition to the exploration of transformer models and their viability for dementia diagnostics, this paper provides a market analysis of a potential automated diagnostics tool utilizing transformer models. The analysis is based on a literature study and on two interviews; one with the CEO of a start-up providing automated dementia tests for healthcare professionals, and one with a psychologist researching dementia as well as potential methods of early diagnosis of dementia. With the interviews and literature study as a basis, we use the SWOT framework, and PEST analysis along with Porter's five forces framework to analyse the current market potential for such an automated tool. Despite detecting several obstacles and difficulties prior to market entry, we find significant potential for such a product given the current state of the market. / Demens är ett syndrom av sjukdomar som orsakar kognitiv nedsättning och påverkar både de drabbade och deras familjer. Den vanligaste typen av demens ar Alzheimers sjukdom, med cirka 10 miljoner nya fall per ar. I denna studie undersöker vi olika maskininlärningsmodeller och tillvägagångssätt i syfte att underlätta får sjukvårdspersonal att ställa en tidig diag­nos, och möjligtvis att även kunna automatisera vissa delar av diagnosprocessen. Vi utvärderar vara modeller på Pitt­corpuset i DementiaBank-datasetet och använder 10-delad ko­rsvalidering. Vi jämför två transformer-modeller: BERT och RoBERTa, och finner att bada modeller åstadkommer goda resultat avseende noggrannhet, precision, specificitet och sen­sitivitet. Den högsta noggrannheten uppnås av RoBERTa, på 86.72%, en precision på 90.69%, och en specificitet på 90.53%. Vidare undersöker vi gångbarheten i att använda automatisk taligenkänning for automatiserad transkribering av ljudinspel­ningar från patientmoten. RoBERTa uppnår da en noggrannhet på 83.59% när den använder transkriberad text från Googles automatiska taligenkänningstjänst, vilket tyder på att sådana metoder kan vara gångbara for att automatisera vissa delar av den diagnostiska processen. Förutom undersökning av transformermodeller bidrar detta verk även med en marknadsanalys av marknadspotentialen får ett verktyg får automatiserad demensdiagnostik. Analysen baseras på en litteraturstudie och två intervjuer; en med en VD får en start-upp som erbjuder liknande tjänster, och en intervju med en forskare inom demens. Med litteraturstudien och de två intervjuerna som grund analyserar vi marknadspotentialen med tre ramverk: Porters fem krafter, PEST-analys och SWOT­analys. Vi fastslår att det trots flertal hinder och svårigheter får marknadsinträde finns det stor potential och en stor efterfrågan på en sådan produkt.

Internal models are believed to be crucial components in human motor control. It has been suggested that the central nervous system (CNS) uses forward and inverse models as internal representations of the motor systems. However, it is still unclear how the CNS implements the high-dimensional control of our movements. In this project, generative adversarial networks (GAN) are studied as a generative model of movement data. It is shown that, for a relatively small number of effectors, it is possible to train a GAN which produces new movement samples that are plausible given a simulator environment. It is believed that these models can be extended to generate high-dimensional movement data. Furthermore, this project investigates the possibility to use a trained GAN as an integrated forward and inverse model for motor control. / Interna modeller tros vara en viktig del av mänsklig rörelsekontroll. Det har föreslagits att det centrala nervsystemet (CNS) använder sig av framåtriktade modeller och inversa modeller för intern representation av motorsystemen. Dock är det fortfarande okänt hur det centrala nervsystemet implementerar denna högdimensionella kontroll. Detta examensarbete undersöker användningen av generativa konkurrerande nätverk som generativ modell av rörelsedata. Experiment visar att dessa nätverk kan tränas till att generera ny rörelsedata av en tvådelad arm och att den genererade datan efterliknar träningsdatan. Vi tror att nätverken även kan modellera mer högdimensionell rörelsedata. I projektet undersöks även användningen av dessa nätverk som en integrerad framåtriktad och invers modell.

Geographic information is one of the most important and the most common types of information in human society. It is estimated that more than 70% of all information in the world has some kind of geographic features. In the era of information explosion, information retrieval (IR) tools, such as search engines, are the main tools people used to quickly find the information they need nowadays. Because of the importance of geographic information, recent efforts have been made either by expanding the traditional IR to support a spatial query, or building a GIR in a brand new architecture from the ground such as the SPIRIT project. To some degree, these existing GIR systems could solve users’ information search need with a spatial filter, especially when the users are looking for information on something within a relatively large extent.Despite its advantage on processing geographical information and queries over conventional IR systems, modern GIR systems are also facing challenges including a proper representation and extraction of geographical information within documents, a better information retrieval model for both thematic and geographical information, a fast indexing mechanism for rapid search within documents by thematic and geographical hints, and even a new architecture of system.The objective of this licentiate research is to provide solutions to some of these problems in order to build a better modern GIR system in the future. The following aspects have been investigated in the thesis: a generic conceptual framework and related key technologies for a modern GIR system, a new information retrieval model and algorithm for measuring the relevance scores between documents and queries in GIR, and finally a new better indexing technique to geographically and thematically index the documents for a faster query processing within modern GIR.Concerning the proposed conceptual framework for modern GIR, it includes three modules: (1) the user interface module, (2) the information extractor, storage and indexer module and (3) the query processing and information retrieval module. Two knowledge bases, Gazetteer and Thesaurus, play an important role in the proposed framework. A digital map based user interface is proposed for the input of user information search needs and representation of retrieval results. Key techniques required for the implementation of a modern GIR using the proposed framework are a proper representation of document and query information, a better geographical information extractor, an innovative information retrieval model and relevance ranking algorithm, and a combined indexing mechanism for both geographical and thematic information.The new information retrieval model is established based on a Spatial Bayesian Network consisting of place names appeared in a single document and the spatial relationships between them. The new model assesses the geographical relevance between GIR document and query by the geographical importance and adjacency of the document geo-footprint versus the geographical scope of the user’s query.Regarding the indexing mechanism for modern GIR systems, a Keyword-Spatial Hybrid Index (KSHI) is proposed for the single and overall geo-footprint model, in which there is only one single geo-footprint for each document to retrieve from. A Keyword-Spatial Dual Index (KSDI) is proved to be more appropriate for a GIR system which allows for multiple geo-footprints within a single document.In addition to theoretical analysis, necessary experiments have also been carried out to evaluate the efficiency of proposed new information retrieval model and indices. Both the theoretical analysis and results of experiments show the potentials of proposed solution and techniques. / QC 20110630

Detta arbete utgår ifrån frågeställningen “Vilka intryck ger olika typer av abstraktion av röstmaterial i en interaktiv ljudinstallation?”. För undersökningen användes en artefakt i form av en interaktiv ljudinstallation som till viss del simulerar en analogradio. Då ljudmaterialet är designat för att vara av skrämmande karaktär fokuserar studien och frågeställningen framförallt på känslan av obehag. Bakgrunden tar upp tidigare exempel på röster i spel, hur ljud kan skapa obehag hos människor, och hur inspelade människoröster respektive syntetiskt genererade röster uppfattas. Undersökningen hade tio deltagare i varierande ålder. Innan testdagen samlades information in om deras ålder och skräckvana. Testet gick ut på att deltagarna fick använda tre versioner av artefakten i fem minuter, och mellan versionerna hölls en intervju gällande vilka intryck som versionen gav. Slutligen ställdes ett par övergripande frågor om deras upplevelse. Resultatet av undersökningen gav inga konkreta svar angående vilka intryck som olika typer av abstraktion ger. Detta främst för att rösterna inte framkom tydligt i ljudbilden, vilket gjorde det svårt för deltagarna att urskilja ord eller typ av röst och därför höra skillnader i förvrängning. Andra faktorer som kan ha bidragit  till detta resultat är testmiljö, deltagarnas skräckvana och buggar i programmet.

Denna undersökning baseras på tidigare forskning som visar på kopplingen mellan förekomsten av stereotyper i dataspel och negativa stereotyper av olika etniska grupper hos spelarna. Bakgrunden beskriver de stereotyper som finns om olika etniska grupper och hur de representeras i digitala spel. Artefakten som användes i studien bestod av en serie karaktärsskisser med olika utseende baserat på länder från olika världsdelar. Utifrån dessa karaktärsskisser besvarade deltagarna enkätfrågor, kring vilka attribut de associerar med de olika karaktärerna. Studiens frågeställning gäller om uppfattningen av dessa karaktärer skiljer sig åt mellan de som spelar mycket och lite dataspel och i så fall på vilket sätt.

Technological advancements made in recent decades in the fields of artificial intelligence (AI) and machine learning (ML) has lead to further automation of tasks previously performed by humans. Manually reviewing and assessing content uploaded to social media and marketplace platforms is one of said tasks that is both tedious and expensive to perform, and could possibly be automated through ML based systems. When introducing ML model predictions to a human decision making process, interpretability and explainability of models has been proven to be important factors for humans to trust in individual sample predictions. This thesis project aims to explore the performance of interpretable ML models used together with humans in an ad review process for a rental marketplace platform. Utilizing the XGBoost framework and SHAP for interpretable ML, a system was built with the ability to score an individual ad and explain the prediction with human readable sentences based on feature importance. The model reached an ROC AUC score of 0.90 and an Average Precision score of 0.64 on a held out test set. An end user survey was conducted which indicated some trust in the model and an appreciation for the local prediction explanations, but low general impact and helpfulness. While most related work focus on model performance, this thesis contributes with a smaller model usability study which can provide grounds for utilizing interpretable ML software in any manual decision making process.