Graduate School of Informatics

This thesis solves the 3D mesh segmentation problem from a different perspective. This perspective contains a data-centric approach with unsupervised learning. This means that 3D mesh segmentation mapping plays a key role in terms of a data-centric approach. We provide a mapping architecture using the node2vec model that also solves the curse of dimension and transforms 3D mesh data into an embedding vector. Segmentation was obtained using the K-Means clustering algorithm using this embedding vector. Additionally, we provide a new strategy that evaluates the graph embedding vector and a new inertia method calculated on 3D mesh data, geodesic inertia.

Date: 02.12.2022 / 14:00 Place: Computer Engineering - A105

Genomic sequencing technologies opened a new era for genetic disorder diagnostics. ​Currently, in over 80% of the cases, the genetic etiology of the diseases can be ​determined by identifying the causative variations. In this study, we have utilized two next-generation sequencing technologies, Ion-Torrent and Illumina, for a rare disease family with two siblings sharing similar symptoms. In addition to comparing technologies, different assemblies of human reference assemblies are analyzed and the benefits of all are discussed for revealing the variants of unsolved rare disease cases. 

Date: 29.11.2022 / 12:30 Place: A108

The main purpose of this thesis is to investigate the potential effects of the gaze bahavior modeling. To that end, experiments were conducted with human participants in the form of a mock-up job interview setting, where a robot avatar programmed with pre-planned gaze behavior in reference to the model was in charge as the interviewer. The eye-tracking data of the participants were collected and analyzed. The eye-tracking results were compared with traditional Human-Robot Interaction survey data along with the TF-IDF (Term Frequency-Inverse Document Frequency) analysis of the post-experimental oral evaluations.

Date: 29.11.2022 / 16:00 Place: B116

In an attempt to understand the semantics of –(y)Iş better, this thesis aims to systematically distinguish imperfect -(y)Iş nominals from perfect -(y)Iş nominals, manner-denoting -(y)Iş nominals from eventuality-denoting ones, identify the types of eventuality denoted, and identify the factivity status of eventuality-denoting -(y)Iş nominals. We suggested sets of tests for each category, and we applied these tests to a sample set of data to demonstrate that they can be used to reliably and accurately make these distinctions.

Date: 17.11.2022 / 11:00 Place: A212

An outcome of taking poor decisions or choosing easier solutions for faster code delivery is technical debt. There is a gap in the field to assist managing technical debt for software development companies that work with software as a service applications. This thesis aims to specify and categorize the technical debt present in organizations using software as a service applications, by studying the processes for categorizing technical debt, and specifying the causes and issues arising due to this debt.

Date: 28.11.2022 / 09:30 Place: A212

The statistical analysis of the microbiome data obtained by the Next Generation Sequencing technology assumes that the data lies on the real space. Since 2017, it has been argued that microbiome data analyzes should be performed on simplex, that is a subspace of real space. Because, microbiome data is "compositional". In this thesis, the characteristics of microbiome data and the statistical difficulties of microbiome analysis are examined. Then, a new grouping method (Principal Microbial Groups) was introduced, using the compositional nature of the data, which is an alternative to the phylogenetic grouping of microbiome data.

Date: 18.11.2022 / 11:30 Place: A212

In deep learning, small input shifts or translations can cause dramatic changes in the output. This is because of the fact that commonly used down-sampling techniques such as max pooling, strided convolution, and average pooling ignore the sampling theorem. We have demonstrated that the cycle-spinning (CS) signal processing technique can be used before down-sampling in deep learning to increase accuracy without introducing any extra learnable parameters. The proposed method can be applied to different algorithms such as GAN, classification, and object detection.

Date: 29.11.2022 / 13:00 Place: METU Informatics Institute