<p>This report explores continuous-time recurrent neural networks (CTRNNs) and their utility in the field of adaptive robotics. The networks herein are evolved in a simulated environment and evaluated on a real robot. The evolved CTRNNs are presented with simple cognitive tasks and the results are analyzed in detail.</p>

<p>The tax history of a company is used to predict corporate bankruptcies using Bayesian inference. Our developed model uses a combination of Naive Bayesian classification and Gaussian Processes. Based on a sample of 1184 companies, we conclude that the Naive Bayes-Gaussian Process model successfully forecasts corporate bankruptcies with high accuracy. A comparison is performed with the current system in place at one of the largest banks in Norway. We present evidence that our classification model, based solely on tax data, is better than the model currently in place.</p>

<p>In this thesis, we have treated the problem of separating content from noise on news websites. We have approached this problem by using TiMBL, a memory-based learning software. We have studied the relevance of the similarity in the training data and the effect of data size in the performance of the extractions.</p>

The purpose of this thesis is to obtain and represent the orientation of the muscle fibres in the left ventricular wall of the human heart. The orientation of these fibres vary continuously through the wall. This report features an introduction to the human heart and medical imaging techniques. Attention is gradually drawn to concepts in computer science, and how they can help us get a “clearer picture” of the internals of, perhaps, the most important organ in the human body. A highly detailed Magnetic Resonance Imaging data set of the left ventricle cavity is used as a base for the analysis with 3-D morphological transformations. Also, a 3-D extension of the Hough transformation is developed. This does not seem to have been done before. An attempt is made to obtain the general trend of the trabeculae carneae, as it is believed that this is the orientation of the inner-most muscle fibres of the heart wall. Suggestions for further work include refinement of the proposed 3-D Hough transformation to yield lines that can be used as guides for parametric curves. Also a brief introduction to Diffusion Tensor Magnetic Resonance Imaging is given.

In this thesis we describe our study of two knowledge intensive Conversational Case-Based Reasoning (CCBR) systems and their methods. We look in particular at the way they have solved inferencing and question ranking. Then we continue with a description of our own design for a CCBR system, that will help patients share their experiences of side effects with drugs, with other patients. We describe how we create cases, how our question selection methods work and present an example of how the domain model will look. It is also included a simulation of how a dialogue would be for a patient. The design we have created is a good basis for implementing a knowledge intensive CCBR system. The system should work better than a normal CCBR system, because of the inferencing and question ranking methods, which should lessen the cognitive load on the user and require fewer questions answered, to reach a good solution.

This paper describes a case based reasoning system that is used to guide the parameters of a segmentation algorithm. Instead of using a fixed set of parameters that gives the best average result over all images, the parameteres are tuned to maximize the score for each image separately. The system's foundation is a set of 20 cases that each contains one 3D MRI image and the parameters needed for its optimal segmentation. When a new image is presented to the system a new case is generated and compared to the other cases based on image similarity. The parameters from the best matching case are then used to segment the new image. The key issue is the use of an iterative approach that lets the system adapt the parameters to suit the new image better, if necessary. Each iteration contains a segmentation and a revision of the result, and this is done until the system approves the result. The revision is based on metadata stored in each case to see if the result has the expected properties as defined by the case. The results show that combining case based reasoning and segmentation can be applied within image processing. This is valid for choosing a good set of starting parameters, and also for using case specific knowledge to guide their adaption. A set of challenges for future research is identified and discussed at length.

This master-thesis considers implementation of automated multimodal volume-to-volume registration of images, in order to provide neurosurgeons with valuable information for planning and intraoperative guidance. Focus has been on medical images from magnetic resonance (MR) and ultrasound (US) for use in surgical guidance. Prototype implementations for MRI-to-US registration have been proposed, and tested, using registration methods available in the Insight Toolkit (ITK). Mattes' Mutual Information has been the similarty metric, based on unpreprocessed angio-graphic volumes from both modalities. Only rigid transformations has been studied, and both types of Gradient Descent and Evolutionary optimizers has been examinated. The applications have been tested on clinical data from relevant surgical operations. The best results were obtained using an evolutional (1+1) optimizer for translational transformations only. This application was both fast and accurate. The other applications, using types of Gradient Descent optimizers, has proved to be significantly slower, inaccurate and more difficult to parameterize. It has been experienced that registration of angio-graphic volumes are easier to accomplish than registration of volumes of other weightings, due to their more similar characteristics. Angio-graphic images are also readily evaluated using volume renderings, but other methods should be constructed to provide a less subjective measure of success for the registration procedures. The obtained results indicate that automatic volume-to-volume registration of angio-graphic images from MRI and US, using Mattes' Mutual Information and an Evolutionary Optimizer, should be feasible for the neuronavigational system considered here, with sufficient accuracy. Further development include parameter-tuning of the applications, to possibly achieve increased accuracy. Additionally, a non-rigid registration application should be developed, to account for local deformations during surgery. Development of additional tools for performing accurate validation of registration results should be developed as well.

Go is an ancient two player board game that has been played for several thousand years. Despite its simple rules, the game requires players to form long-term strategic plans and also possess strong tactical skills to handle the complex fights that often occur during a game. From an artificial intelligence point of view, Go is notable as a game that has been highly resistant to all traditional game playing approaches. In contrast to other board games such as chess and checkers, top human Go players are still significantly better than any computer Go playing programs. It is believed that the strategic depth of Go will require the use of new and more powerful artificial intelligence methods than the ones successfully used to create computer players for such other games. There have been some promising new developments using new Monte Carlo-based techniques to play computer Go in recent years, and programs based on this approach are currently the strongest computer Go players in the world. However, even these programs still play at an amateur level, and they cannot compete with professional or strong amateur human players. In this thesis we explore the idea of reusing experience from previous games to identify strategically important moves for a Go board position. This is based on finding a previous game position that is highly similar to the one in the current game. The moves that were played in this previous game are then adapted to generate new moves for the current game situation. A new computer Go playing system using Monte Carlo-based Go methods was designed as a part of this thesis work, and a prototype implementation of this system was also developed. We extended this initial prototype using case based reasoning (CBR) methods to quickly identify the most strategically valuable areas of the board at the early stages of the game, based on finding similar positions in a collection of professionally played games. The last part of the thesis is an evaluation of the developed system and the results observed using our implementation. These results show that our CBR-based approach is a significant improvement over the initial prototype, and in the opening game it allows the program to quickly locate the most strategically interesting areas of the board. However, by itself our approach does not find strong tactical moves within these identified areas, and thus it is most valuable when used to provide strategic guidelines for other methods that can find tactical plays.

In this project we seek to explore whether it is possible for an artificial agent to learn how to play curling. To achieve this goal we developed a simulator that works as an environment where different agents can be tested against each other. Our most successful agent use a Linear Target Function as a basis for selecting good moves in the game. This agent has become very adept at placing stones, but we discovered that it lacks the ability to employ advanced strategies that reach over more than just one stone. In an effort to give the agent this ability we expanded it using Q-learning with UCT, however this was not successful. For the agent to work we need a good representation of the information in curling, and our representation was quite broad. This caused the training of the agent to take an unreasonably large amount of time.

The gameplay of real-time strategy games can be divided into macromanagement and micromanagement. Several researchers have studied automated learning for macromanagement, using a case-based reasoning/reinforcement learning architecture to defeat both static and dynamic opponents. Unlike the previous research, we present the Unit Priority Artificial Intelligence (UPAI). UPAI is a case-based reasoning/reinforcement learning system for learning the micromanagement task of prioritizing which enemy units to attack in different game situations, through unsupervised learning from experience. We discuss different case representations, as well as the exploration vs exploitation aspect of reinforcement learning in UPAI. Our research demonstrates that UPAI can learn to improve its micromanagement decisions, by defeating both static and dynamic opponents in a micromanagement setting.