In most modern video games, character behavior is scripted. No matter how many times the player exploits a weakness, that weakness is never repaired. To create realistic artificial intelligence in games, it is important to understand how people reason in game like environments and how they arrive at their decisions. This project examines if environmental factors affect player’s decisions, when it comes to completing a specific task in a controlled game like scenario. The project’s empirical approach with experiments, confirms that definitely some environmental factors have implications on the human decision making in gaming environments. This knowledge could lead to a better gaming experience and new genres of games. / I de flesta moderna TV-spel, är de artificiella spelkaraktärernas beteenden förutbestämda. Oavsett hur många gånger en spelare utnyttjar en svaghet, så repareras denna aldrig. För att skapa realistisk artificiell intelligens i spel, krävs det förståelse om hur människor resonerar i spelmiljöer och hur de kommer fram till sina beslut. Detta projekt undersöker hur miljöfaktorer påverkar spelares beslut när det gäller att slutföra en viss uppgift i ett kontrollerat spelliknande scenario. Projektets empiriska tillvägagångssätt med experiment, bekräftar att det definitivt existerar miljöfaktorer som har konsekvenser för det mänskliga beslutsfattandet i spelmiljöer. Denna kunskap kan bidra till en bättre spelupplevelse och nya genrer av spel.

Some machine learning tasks have a complex output, rather than a real number or a class. Those outputs are composed by elements which have interdependences and structural properties. Methods which take into account the form of the output are known as structured prediction techniques. This study focuses on those techniques, evaluating their performance for tasks of sequence labeling and comparing them. Specifically, tasks of natural language processing are used as benchmarks.  The principal problem evaluated is part-of-speech tagging. Datasets of different languages (English, Spanish, Portuguese and Dutch) and environments (newspapers, twitter and chats) are used for a general analysis. Shallow parsing and named entity recognition are also examined. The algorithms treated are structured perceptron, conditional random fields, structured support vector machines and trigram hidden Markov models. They are also compared to different approaches to solve these problems. The results show that, in general, structured perceptron has the best performance for sequence labeling with the conditions evaluated. However, with few training examples, structured support vector machines can achieve a similar or superior accuracy. Moreover, the results for conditional ranom fields is near those two methods. The relative results of the algorithms are similar across different datasets, but the absolute accuracies are dependent on their specificities.

The purpose of this research is to test whether clustering algorithmscan be used to detect troll farms in social networks. Troll farms are profes-sional organizations that spread disinformation online via fake personas.The research involves a comparative study of two different clustering algo-rithms and a dataset of Twitter users and posts that includes a fabricatedtroll farm. By comparing the results and the implementations of the K-means as well as the DBSCAN algorithm we have concluded that clusteranalysis can be used to detect troll farms and that DBSCAN is bettersuited for this particular problem compared to K-means. / Målet med denna rapport är att testa om klusteringalgoritmer kananvändas för att identifiera trollfarmer på sociala medier. Trollfarmer ärprofessionella organisationer som sprider desinformation online med hjälpav falska identiteter. Denna rapport är en jämförande studie med två olikaklusteringalgoritmer och en datamängd av Twitteranvändare och tweetssom inkluderar en fabrikerad trollfarm. Genom att jämföra resultaten ochimplementationerna av algoritmerna K-means och DBSCAN får vi framslutsatsen att klusteralgoritmer kan användas för att identifiera trollfar-mar och att DBSCAN är bättre lämpad för detta problem till skillnadfrån K-means.

Natural language interfaces aim at providing seamless interaction for users to the functionality computer systems offer. Much of this functionality is accessible online through web based application programming interfaces. A NLI capable of leveraging this has the potential of being a powerful tool and to our knowledge, no previous research in this area exist. Therefore, this thesis aims at investigating how existing methods and concepts can be applied to the subject of a NLI over several APIs. The thesis is based on an extensive literature study focused on state-of-the-art within portability of NLI research. Methods from this are discussed in the context of a NLI over APIs and a suitable way of approaching a solution is presented. The approach divides the problem into identifying a domain for the conversation using available knowledge bases and handling specific domains individually using dialog state trackers. The main finding of the study is that it is feasible to create a scalable general purpose NLI over web APIs leveraging existing research. / Gränssnitt för naturligt språk ämnar att underlätta interaktionen för användare till den funktionalitet som datorsystem erbjuder. Mycket av denna funktionalitet finns tillgänglig online genom webb-baserade applikationsprogrammeringsgränssnitt. Ett NLI kapabelt att utnyttja denna funktionalitet har potentialen att vara ett kraftfullt verktyg och till vår vetskap har ingen forskning gjorts inom detta område. På denna grund ämnar denna undersökning att finna hur nuvarande metoder och koncept kan appliceras till ett NLI för flera APIer. Denna rapport bygger på en bred literaturundersökning fokuserad på den mest aktuella forskningen inom portabilitet för NLI. Metoder från detta område diskuteras i sammanhanget av ett NLI över APIer och ett lämpligt sätt att närma sig en lösning presenteras. Lösningen delar upp problemet i att identifiera domän för konversationen genom att använda tillgängliga kunskapsbaser samt att använda “dialog state trackers” för att hantera individuella domäner. Den huvudsakliga slutsatsen som dras är att det är möjligt att skapa ett skalbart NLI kapabelt att hantera flera APIer genom att applicera nuvarande forskning.

This study has investigated the suitability of the eye-tracker X2-30 from Tobii when measuring a student’s alertness level. To be able to determine the suitability of the X2-30 two different environments and two different subjects has been used. The environments were a lecture and webinars, and the two subjects were the two male authors of this study. The results indicate that the eye-tracker X2-30 can be useful when it’s used under optimal circumstances (i.e. webinars), since it needs to have few disturbing factors when gathering data. These factors regard the subjects not moving out of the tracking area, i.e. the area of which the X2-30 is able to track the eye, and consequently it also means that the subject needs to be careful not to move the hardware. When used under optimal circumstances it was possible to obtain accurate data regarding pupil behaviour and somewhat accurate data regarding blink rate, making data relevant and the equipment suitable for measuring of alertness under the correct circumstances. / Denna studie undersöker lämpligheten hos Tobii ABs X2-30 eye-tracker för att mäta en students vakenhet. För att avgöra lämpligheten så har tester gjorts i två olika miljöer och med två olika försökspersoner. Miljöerna var en föreläsning och ett flertal webinars medan försökspersonerna var de två manliga författarna av denna studie. Resultaten indikerar att X2-30 kan vara användbar när den används under optimala omständigheter, det vill säga webinars, då den kräver så få störande faktorer som möjligt. Dessa faktorer rör utrustningens mätzon och att försökspersonen inte lämnar denna, samt att hårdvaran inte flyttas under mätning. När utrustningen användes under optimala förhållanden så var det möjligt att få precis data rörande pupillens beteende och ganska korrekt data rörande blinkfrekvens vilket gör utrustningen lämplig för mätning av vakenhet under rätt omständigheter.

Epilepsy affects nearly 1% of the world’s population and is a disorder that affects the central nervous system where the nerve cell activity in the brain becomes disrupted and causes seizure. Further study in this field can minimize the potential damage caused by individuals with epilepsy. This thesis investigated the the optimal window size for classification of EEG epochs as scientists today simply guess a window size and therefore might not get the best possible results. The method that was used was an existing software where the window size could be easily changed and investigated. The software used Support Vector Machines and calculated the probability for seizures. The results from this investigation showed more fluctuating probabilities at lower and higher window sizes, and more stable at 60 seconds to 180 seconds. In conclusion, the optimal window size can be argued to be around 90 seconds as it has the highest maximum, minimum, and average probability.

With interdisciplinary research being a possibility in modern research environ- ments, it is interesting to optimise collaborations between researchers in order to further develop the research environment. The scope of this thesis was therefore to develop a method to measure how widespread the interdisciplinary research is and to propose collaboration groups of researchers created by the use of graph theory. This problem was approached by studying the research at KTH by col- lecting research publications from the publication database DiVA using a web crawler. Then representing the authors of the publications as nodes and the collaborations between two authors as edges in a graph. A graph partitioning algorithm developed by Flake et al. was chosen after a literature study, then applied to the graph to produce the requested collaboration groups. The results showed that while interdisciplinary research is not the norm at KTH, 23% of the proposed collaboration groups consisted of two or more researchers from different schools at KTH. The original ratio of school associ- ation was retained through the partitioning of the graph. A measurement of collaboration per researcher in each collaboration group was suggested and the calculated values of these measurements was found to be largely in the same range, with the exception of one collaboration group. The results also high- lighted some inconsistencies in DiVA. The conclusions were that interdisciplinary research was not very widespread at KTH, however 77 groups were suggested which could be of use for researchers at KTH from now on and in the future. A conclusion was also that this method for finding suitable collaboration groups could be applied at other universities where perhaps interdisciplinary research is more frequent.

Setting up online contests with gamification is an effective marketing method, but which brings security complications. By having rewards with high value, cheaters are attracted to participate with the use of malicious bots. To distinguish bots from humans different methods are used which are divided into Human Interactive Proof (HIP) and Human Observational Proof (HOP). This report aims to look at the effectiveness of the most popular HIPs and HOPs and how an attacker is able to bypass them. From the results, parameters that are of interest when implementing a framework to detect and prevent malicious bots, are presented. Data was collected from five honeypot systems. It is concluded that CAPTCHAs should be used as much as possible, together with HMAC and an Intrusion Detection System (IDS) based on click diversity and submissions per IP-address.

Accurate prediction of future stock market prices is of great importance to traders. The process can be automated using articial neural networks. However, the conventional backward propagation of errors algorithm commonly used for training the networks suffers from the local minima problem. This study investigates whether investing more computational resources into training an ar-ticial neural network using genetic algorithms over the conventional algorithm,to avoid the local minima problem, can result in higher prediction accuracy. The results indicate that there is no signicant increase in accuracy to gain by investing resources into training with genetic algorithms, using our proposed model.

In this report, genetic algorithms were compared to BFS and DFS for solvingmazes with regards to computational time and solution path length. This wasdone by creating mazes with Wilson's algorithm and running several tests ina controlled environment. The solution path length found by the genetic algorithmwas in general better than the DFS and often almost as good as the onefound by the BFS. The computational time needed by the genetic algorithmhowever was signicantly worse than the time needed by both the DFS and theBFS. / I denna rapport har genetiska algoritmer jamforts med BFS och DFS for attlosa labyrinter med avseende pa berakningstid och losningslangd. Detta gjordesgenom att skapa labyrinter med Wilsons algoritm och kora en stor mangd testeri en kontrollerad miljo. Losningslangderna som den genetiska algoritmen fannvar i allmanhet battre an de av DFS och ofta nastan lika bra som de som BFSfann. Den genetiska algoritmens berakningstid var dock betydligt langre an dentid som kravdes for bade DFS och BFS.