This thesis is deal with classification for retinal nerve fibre layer. Texture features from six texture analysis methods are used for classification. All methods calculate feature vector from inputs images. This feature vector is characterized for every cluster (class). Classification is realized by three supervised learning algorithms and one unsupervised learning algorithm. The first testing algorithm is called Ho-Kashyap. The next is Bayess classifier NDDF (Normal Density Discriminant Function). The third is the Nearest Neighbor algorithm k-NN and the last tested classifier is algorithm K-means, which belongs to clustering. For better compactness of this thesis, three methods for selection of training patterns in supervised learning algorithms are implemented. The methods are based on Repeated Random Subsampling Cross Validation, K-Fold Cross Validation and Leave One Out Cross Validation algorithms. All algorithms are quantitatively compared in the sense of classication error evaluation.

The thesis advanced image segmentation for 3D image deals with segmentation and anaglyph 3D views. In the theoretical part of the thesis describes the different approaches were used to image segmentation and closely related methods of image processing. In the following practical part was the implementation of selected methods and created user-friendly applications. The main objective of the program is to identify significant objects in the image. For the purpose of segmentation methods have been implemented based on k-means method, the method of contour and the growth of seeds. The program is created in Visual Studio 2008 and written in C + +. The input and output is the image in various formats (JPG, BMP, TIFF).

This thesis describes the use of NetFlow data in the systems for detection of disruptions or anomalies in computer network traffic. Various methods for network data collection are described, focusing especially on the NetFlow protocol. Further, various methods for anomaly detection  in network traffic are discussed and evaluated, and their advantages as well as disadvantages are listed. Based on this analysis one method is chosen. Further, test data set is analyzed using the method. Algorithm for real-time network traffic anomaly detection is designed based on the analysis outcomes. This method was chosen mainly because it enables detection of anomalies even in an unlabelled network traffic. The last part of the thesis describes implementation of the  algorithm, as well as experiments performed using the resulting  application on real NetFlow data.

In der Epigenetik werden die Veränderungen der Erbinformationen neben der DNS erforscht. Dabei werden den Histonen, um die sich die DNS im Zellkern wickelt, eine große Bedeutung zugeordnet. In dieser Arbeit werden die Ergebnisse eines neuen Segmentierungsverfahrens ausgewertet und visualisiert. Dabei werden die vorliegenden Daten mittels des k-means++ Algorithmus geclustert.Zuerst werden die Clusterergebnisse statistisch ausgewertet, um sie dann mit den durch vorgehenden Arbeiten erworbenen Kenntnissen zu vergleichen. Mittels dieses Vergleichs werden dann die idealen Parameter für das Clustering bestimmt. Die Ergebnisse dieses idealen Clusterings werden dann mittels Starplots, Scatterplots und Binningplots visualisiert. Für die Erstellung der Scatter- und Binningplots wird eine PCA genutzt, um die Daten auf zwei Dimensionen zu reduzieren.

Fortnox takes in many errands via their support chat. Some of the questions can be hard to interpret, making it difficult to know where to delegate the question further. It would be beneficial if the process was automated to answer the questions instead of need to put in time to analyze the questions to be able to delegate them. So, the main task is to find an unsupervised model that can take questions and put them into topics. A literature review over NLP and clustering was needed to find the most suitable models and techniques for the problem. Then implementing the models and techniques and evaluating them using support chat questions received by Fortnox. The unsupervised models tested in this thesis were LDA and K-means. The resulting models after training are analyzed, and some of the clusters are given a label. The authors of the thesis give clusters a label after analyzing them by looking at the most relevant words for the cluster. Three different sets of labels are analyzed and tested. The models are evaluated using five different score metrics: Silhouette, AdjustedRand Index, Recall, Precision, and F1 score. K-means scores the best when looking at the score metrics and have an F1 score of 0.417. But can not handle very small documents. LDA does not perform very well and got i F1 score of 0.137 and is not able to categorize documents together.

The study of enzyme binding sites is an essential but rather demanding process of increased complexity since the amino acids lining these areas are not rigid. At the same time, the minimization of side effects and the specificity of new ligands is a great challenge in the structure-based drug design approach. Using glycogen phosphorylase - a validated target for the development of new antidiabetic agents - as a case study, this project focuses on the examination of side-chain conformations of amino acids that play a key role in the catalytic site of the enzyme. Specifically, different rotamers of each amino acid were collected to build a dataset of different conformations of the catalytic site. The rotamers were filtered by their probability of occurrence and subsequently, all rotamers that create steric clashes were rejected. Then, these conformations were clustered based on their similarity. Three different clustering algorithms and multiple numbers of clusters were tested using the silhouette scores evaluation for the clustering process. In order to measure the similarity, the Euclidean metric was used which due to the correspondence of the coordinates between the conformations was very similar to the cRMSD metric. Two-level clustering was applied to the dataset for more in-depth observations. According to the clustering results, specific aminoacids with major geometrical variations in their rotamers play the most important role in the separation of the clusters. Additionally, all rotamers of an amino acid can be grouped based on their structure, something that was confirmed using “Chimera” software as a visualization tool. To this end, the ultimate aim of this study is to examine whether the clustering of conformations produces clusters with points geometrically similar to each other, in order to identify near neighbors, i.e. conformations that are quite similar in structure but do not play a determinant role in the function and those that are quite diverse and could be further exploited.

There are common pitfalls and neglected areas when using clustering approaches to solve educational problems. A clustering algorithm is often used without the choice being justified. Few comparisons between a selected algorithm and a competing algorithm are presented, and results are presented without validation. Lastly, few studies fully utilize data provided in an educational environment to evaluate their findings. In response to these problems, this thesis describes a rigorous study comparing two clustering algorithms in the context of an educational digital library service, called the Instructional Architect. First, a detailed description of the chosen clustering algorithm, namely, latent class analysis (LCA), is presented. Second, three kinds of preprocessed data are separately applied to both the selected algorithm and a competing algorithm, namely, K-means algorithm. Third, a series of comprehensive evaluations on four aspects of each clustering result, i.e., intra-cluster and inter-cluster distances, Davies-Bouldin index, users' demographic profile, and cluster evolution, are conducted to compare the clustering results of LCA and K-means algorithms. Evaluation results show that LCA outperforms K-means in producing consistent clustering results at different settings, finding compact clusters, and finding connections between users' teaching experience and their effectiveness in using the IA. The implication, contributions, and limitation of this research are discussed.

京都大学 / 0048 / 新制・課程博士 / 博士(医学) / 甲第19616号 / 医博第4123号 / 新制||医||1015(附属図書館) / 32652 / 京都大学大学院医学研究科医学専攻 / (主査)教授 佐藤 俊哉, 教授 富樫 かおり, 教授 藤渕 航 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM

Information and communications technology (ICT) is an important part in today’s society and around 60% of the world's population are connected to the internet. Processing and storing ICT data corresponds to approximately 1% of the global electricity demand. Locations that store ICT data produce a lot of heat that needs to be cooled, and the cooling systems stand for up to 40% of the total energy used in ICT-centre locations. Investigating the efficiency of the cooling in ICT-centres is important to make the whole ICT-centre more energy efficient, and possibly saving operational costs. Unwanted operational behaviour in the cooling system can be analysed by using unsupervised machine learning and clustering of data. The purpose of this thesis is to characterise cooling patterns, using K-means clustering, in two water-cooled ICT rooms. The rooms are located at Ericsson’s facilities in Linköping Sweden. This will be fulfilled answering the research questions: RQ1. What is the cooling power per m2 delivered by the cooling equipment in the two different ICT rooms at Ericsson?  RQ2. What operational patterns can be found using a suitable clustering algorithm to process and compare data for LCP at two ICT-rooms?   RQ3. Based on information from RQ1 and patterns from RQ2 what undesired operational behaviours can be identified for the cooling system? The K-means clustering is applied to time series data collected during the year of 2022 which include temperatures of water and air; electric power and cooling power; as well as waterflow in the system. The two rooms use Liquid Cooling Packages (LCP)s, also known as in-row cooling units, and room 1 (R1) also include computer room air handlers (CRAHs). K-means clusters each observation into a group that share characteristics and represent different operating scenarios. The elbow-method is used to determine the number of clusters, it created four clusters for R1 and three clusters for room 2 (R2).  Results show that the operational patterns differ between R1 and R2. The cooling power produced per m2 is 1.36 kW/m2 for R1 and 2.14 kW/m2 for R2. Cooling power per m3 is 0.39 kW/m3 for R1 and 0.61 kW/m3 for R2. Undesirable operational behaviours were identified through clustering and visual representation of the data. Some LCPs operate very differently even when sharing the same hot aisle. There are disturbances such as air flow and setpoints that create these differences, which results in that some LCPs operate with high cooling power and others that operate with low cooling power. The cluster with the highest cooling power is cluster 4 and 3 for R1 and R2 respectively. Cluster 2 has the lowest cooling power in R1 and R2. For LCPs operating in cluster 2 where waterflow mostly at 0 l/min and therefore where not contributing to the cooling of the rooms. Lastly, the supplied electrical power and produced cooling power match in R1 but do not in R2. Implying that heat leave the rooms by other means than via the cooling system or faulty measurements. There is a possibility to investigate this further. Water in R1 and R2 is found to, at occasions, exit the room with temperature below the ambient room temperature. It is also concluded that the method functions to identify unwanted operational behaviours, knowledge that can be used to improve ICT operations.  To summarize, undesired operational behaviours can be identified using the unsupervised machine learning technique K-means clustering.

This thesis investigates how various driver signals, signals that a truck driver can interact with, influences fuel consumption and what are the optimal values of these signals in various traffic conditions. More specifically, the objective is to estimate good driver behavior in various traffic conditions and compare bad driver behavior in similar situations to see how performing a specific driver action, changing a driver signal from the bad driver value to the corresponding good driver value impacts the fuel consumption. The result is an AI-based algorithm that utilizes the transformer model architecture to estimate good driver behavior, based on environmental describing signals, as well as fuel consumption. Utilizing these, causal inference is used to estimate how much fuel can be saved by switching a driver signal from a bad driver value to a good driver value.