This thesis compares forward neural networks with algorithms using game theory on basis of board game Blokus. The theoretical introduction part describes the characteristics of neural networks and work with them. There is also outlined algorithm of game theory. The second part deals about the implementation of players based on the outlined principles  and shortly descriptions GUI of application. In conclusion, the differences between the players  are evaluated on the charts created on the performed tests.

No description available.

Vehicle acoustic signals have long been considered as unwanted traffic noise. In this research acoustic signals generated by each vehicle will be used to detect its presence and classify its type. Circular arrays of microphones were designed and built to detect desired signals and suppress unwanted ones. Circular arrays with multiple rings have an interesting and important property that is constant sidelobe levels. A modified genetic algorithm that can work directly with real numbers is used in the circular array design. It offers more effective ways to solve numerical problems than a standard genetic algorithm. In classifier design two main paradigms are considered: multilayer perceptrons and adaptive fuzzy logic systems. A multilayer perceptron is a network inspired by biological neural systems. Even though it is far from a biological system, it possesses the capability to solve many interesting problems in variety fields. Fuzzy logic systems, on the other hand, were inspired by human capabilities to deal with fuzzy terms. Its structures and operations are based on fuzzy set theory and its operations. Adaptive fuzzy logic systems are fuzzy logic systems equipped with training algorithms so that its rules can be extracted or modified from available numerical data similar to neural networks. Both fuzzy logic systems and multilayer perceptrons have been proved to be universal function approximators. Since there are approximations in almost every stage, both of these system types are good candidates for classification systems. In classification problems unequal learning of each class is normally encountered. This unequal learning may come from different learning difficulties and/or unequal numbers of training data from each class. The classifier tends to classify better for a well-learned class while doing poorly for other classes. Classification costs that may be different from class to class can be used to train and test a classifier. An error backpropagation algorithm can be modified so that the classification costs along with unequal learning factors can be used to control classifier learning during its training phase. / Ph. D.

The study of the instabilities in boiling water reactors is of significant importance to the safety withwhich they can be operated, as they can cause damage to the reactor posing risks to both equipmentand personnel. The instabilities that concern this paper are progressive growths in the oscillatingpower of boiling-water reactors. As thermal power is oscillatory is important to be able to identifywhether or not the power amplitude is stable. The main focus of this paper has been the development of a neural network estimator of these insta-bilities, fitting a non-linear model function to data by estimating it’s parameters. In doing this, theambition was to optimize the networks to the point that it can deliver near ”best-guess” estimationsof the parameters which define these instabilities, evaluating the usefulness of these networks whenapplied to problems like this. The goal was to design both MLP(Multi-Layer Perceptron) and SVR/KRR(Support Vector Regres-sion/Kernel Rigde Regression) networks and improve them to the point that they provide reliableand useful information about the waves in question. This goal was accomplished only in part asthe SVR/KRR networks proved to have some difficulty in ascertaining the phase shift of the waves.Overall, however, these networks prove very useful in this kind of task, succeeding with a reasonabledegree of confidence to calculating the different parameters of the waves studied.

Machine learning is a field that is inspired by how humans and, by extension, the brain learns.The brain consists of a biological neural network that has neurons that are either active or inactive. Modern-day artificial intelligence is loosely based on how biological neural networks function. This paper investigates whether a multi layered perceptron that utilizes inactive/active neurons can reduce the number of active neurons during the forward and backward pass while maintaining accuracy. This is done by implementing a multi layer perceptron using a python environment and building a neuron activation algorithm on top of it. Results show that it ispossible to reduce the number of active neurons by around 30% with a negligible impact on test accuracy. Future works include algorithmic improvements and further testing if it is possible to reduce the total amount of mathematical operations in other neural network architectures with a bigger computational overhead.

At present day, distributed computing is a widely used technique, where volunteers support different computing power needs organizations might have. This thesis sought to benchmark distributed computing performance limited to mobile device support since this type of support is seldom done with mobile devices. This thesis proposes two approaches to harnessing computational power and infrastructure of a group of mobile devices. The problems used for benchmarking are small instances of deep learning training. One requirement posed by the mobile devices’ non-static nature was that this should be possible without any significant prior configuration. The protocol used for communication was HTTP. The reason deep-learning was chosen as the benchmarking problem is due to its versatility and variability. The results showed that this technique can be applied successfully to some types of problem instances, and that the two proposed approaches also favour different problem instances. The highest request rate found for the prototype with a 99% response rate was a 2100% increase in efficiency compared to a regular server. This was under the premise that it was provided just below 2000 mobile devices for only particular problem instances. / För närvarande är distribuerad databehandling en utbredd teknik, där frivilliga individer stödjer olika organisationers behov av datorkraft. Denna rapport försökte jämföra prestandan för distribuerad databehandling begränsad till enbart stöd av mobila enheter då denna typ av stöd sällan görs med mobila enheter. Rapporten föreslår två sätt att utnyttja beräkningskraft och infrastruktur för en grupp mobila enheter. De problem som används för benchmarking är små exempel på deep-learning. Ett krav som ställdes av mobilenheternas icke-statiska natur var att detta skulle vara möjligt utan några betydande konfigureringar. Protokollet som användes för kommunikation var HTTP. Anledningen till att deeplearning valdes som referensproblem beror på dess mångsidighet och variation. Resultaten visade att denna teknik kan tillämpas framgångsrikt på vissa typer av probleminstanser, och att de två föreslagna tillvägagångssätten också gynnar olika probleminstanser. Den högsta requesthastigheten hittad för prototypen med 99% svarsfrekvens var en 2100% ökning av effektiviteten jämfört med en vanlig server. Detta givet strax under 2000 mobila enheter för vissa speciella probleminstanser.

The term Rogue Trading is defined as the activity of someone at a financial organisation losing a large amount of money in bad or illegal transactions and trying to hide this. The activity of Rogue traders exposes financial organisations to huge risks and may lead to the organisation collapsing, which will affect other stakeholders like, for example, the customers. In order to detect potential Rogue Trading cases, Control Systems that monitor the employees and the positions they take on financial markets must exist. In this study, a two-step control system is suggested to monitor the margins on Foreign exchange (FX) Forwards traded by employees at the Swedish bank Skandinaviska Enskilda Banken (SEB). The first step in the control system uses a Deep Learning neural network trained on transactional data to predict the margin. The errors of the predictions versus the actual values are then in the second step of the control system used to find outliers which should be flagged for further investigation due to a too high deviation. The results show that the model hopefully can decrease the number of false positives yielded by the current Control Systems at SEB and thus reduce manual inspection of flagged transactions. / Termen Rouge Trading definieras som en aktivitet där någon på en finansiell institution förlorar stora mängder pengar i dåliga eller illegala transaktioner och försöker dölja detta. Detta är något som skapar enorma risker för finansiella institutioner och som kan förorsaka organisationens kollaps, som kan påverka intressenter som till exempel kunder. För att upptäcka potentiella företeelser av Rouge Trading så måste kontrollsystem som övervakar anställda och deras positioner existera. I denna studie föreslås och presenteras ett tvåstegs-system för att övervaka marginaler vid terminsaffärer i utländsk valuta vid Skandinaviska Enskilda Banken (SEB). Det första steget i kontrollsystemet använder ett neuralt närverk tränat på data från transaktioner för att prediktera en marginal. Differenserna mellan prediktionen och det faktiska värdet används för att finna outliers vilka borde flaggas för vidare undersökning. Resultaten visar att modellen förhoppningsvis kan minska antalet falska positiva som det nuvarande kontrollsystemet ger på SEB, något som således kan minska den manuella inspektionen av flaggade transaktioner.

No description available.

No description available.

L'extraction de structures logiques de documents est un défi du fait de leur complexité inhérente et du fossé existant entre les observations extraites de l'image et leur interprétation logique. La majorité des approches proposées par la littérature sont dirigées par le modèle et ne proposent pas de solution générique pour des documents complexes et bruités. Il n'y a pas de modélisation ni d'explication sur les liens permettant de mettre en relation les blocs physiques et les étiquettes logiques correspondantes. L'objectif de la thèse est de développer une méthode hybride, à la fois dirigée par les données et par le modèle appris, capable d'apprentissage et de simuler la perception humaine pour effectuer la tâche de reconnaissance logique. Nous avons proposé le Réseau de Neurones Dynamique Perceptif qui permet de s'affranchir des principales limitations rencontrées dans les précédentes approches. Quatre points principaux ont été développés : - utilisation d'une architecture neuronale basée sur une représentation locale permettant d'intégrer de la connaissance à l'intérieur du réseau. La décomposition de l'interprétation est dépliée à travers les couches du réseau et un apprentissage a été proposé pour déterminer l'intensité des liaisons ; - des cycles perceptifs, composés de processus ascendants et descendants, accomplissent la reconnaissance. Le réseau est capable de générer des hypothèses, de les valider et de détecter les formes ambigües. Un retour de contexte est utilisé pour corriger les entrées et améliorer la reconnaissance ; - un partitionnement de l'espace d'entrée accélérant la reconnaissance. Des sous-ensembles de variables sont créés automatiquement pour alimenter progressivement le réseau afin d'adapter la quantité de travail à fournir en fonction de la complexité de la forme à reconnaître ; - l'intégration de la composante temporelle dans le réseau permettant l'intégration de l'information de correction pendant l'apprentissage afin de réaliser une reconnaissance plus adéquate. L'utilisation d'un réseau à décalage temporel permet de tenir compte de la variation des entrées après chaque cycle perceptif tout en ayant un fonctionnement très proche de la version statique.