| 1 |
| Title |
Financial Report Summarization |
| Speaker |
Abdullah Rauf Şimşek |
| Date |
December 18th, 2024 |
| Time |
13:30 (GMT +3) |
| Advisor |
Ahmet Cüneyd Tantuğ |
Abstract
Financial reports are packed with important information, but they can be long and difficult to read. Summarizing these reports using natural language processing helps make the key points more accessible and easier to understand. The goal is to create summaries that capture the most important insights, like financial performance and trends, while skipping over the extra details. In low-resource settings, where there may not be enough data, data augmentation techniques can be used to create more training examples for the model, improving its ability to generate accurate summaries. By automating this process, we can save time for investors, analysts, and anyone who needs a quick overview of a company’s status.
|
| Location |
Informatics Institute
Room 317
|
|
| 2 |
| Title |
Identifitacion of Alzheimer’s Disease by Using Network Analysis |
| Speaker |
Muhammed Ömer Taylan |
| Date |
December 18th, 2024 |
| Time |
14:00 (GMT +3) |
| Advisor |
Sefer Baday |
Abstract
This research presents an advanced investigation into the genetic architecture of Alzheimer's disease, a progressive neurodegenerative disorder marked by cognitive decline and dementia.
Employing Weighted Gene Co Expression Network Analysis and graph theory algorithms, the study analyzes publicly accessible microarray data to unravel complex gene expression networks in AD. Thirteen distinct gene modules are identified, and their correlation with clinical data is meticulously examined. Utilizing graph centrality algorithms, the study highlights key genes within these modules, offering new insights into the genetic dynamics of Alzheimer's disease. This approach not only deepens our understanding of the molecular basis of AD but also has the potential to identify novel therapeutic targets, addressing the urgent need for more effective treatments in light of the disease's increasing prevalence and substantial societal impact.
|
| Location |
Informatics Institute
Room 317
|
|
| 3 |
| Title |
Autoencoders for Hyperspectral Imagery Data |
| Speaker |
Ufuk Kaya |
| Date |
December 18th, 2024 |
| Time |
14:30 (GMT +3) |
| Advisor |
Behçet Uğur Töreyin |
Abstract
Hyperspectral imagery (HSI) captures a vast range of spectral information across numerous bands, making it invaluable for applications in remote sensing, environmental monitoring, and material identification. However, the high dimensionality and complexity of HSI data pose significant challenges for efficient analysis and processing. This seminar explores the use of autoencoders, a deep learning-based unsupervised learning technique, for dimensionality reduction and feature extraction in hyperspectral imagery. The autoencoder's ability to compress high-dimensional input data into a lower-dimensional latent space and reconstruct it with minimal loss of information makes it a powerful tool for HSI data processing.
|
| Location |
Informatics Institute
Room 317
|
|
| 4 |
| Title |
Early Diagnosis of Parkinson’s Disease by Using Graph Neural Networks Based on Patient Similarity Networks |
| Speaker |
Doğu Berk Gümüş |
| Date |
December 18th, 2024 |
| Time |
15:00 (GMT +3) |
| Advisor |
Sefer Baday |
Abstract
Parkinson Disease (PD) is a neurodegenerative disorder affecting motor and non-motor functions, leading to decreased quality of life and life expectancy. Diagnosis is challenging due to the absence of definitive biomarkers and the varied presentation of symptoms. This study utilizes patient similarity networks with Graph Neural Networks (GNNs) to improve PD diagnosis. Using the Parkinson’s Progression Markers Initiative (PPMI) dataset, a network is constructed based on genetic markers and symptom profiles to classify subjects into PD, Prodromal, and Healthy Controls. The findings demonstrate that GNNs can enhance diagnostic precision by leveraging comprehensive patient profiles, enabling earlier intervention and better patient outcomes.
|
| Location |
Informatics Institute
Room 317
|
|
| 5 |
| Title |
A Comparative Study of AODV, DSR, and DYMO Protocols in Mobile Ad-Hoc Networks across Multiple Scenarios |
| Speaker |
Mihriye Tekir |
| Date |
December 25th, 2024 |
| Time |
13:30 (GMT +3) |
| Advisor |
Altan Çakır |
Abstract
Mobile Ad-Hoc Networks (MANETs) depend on effective routing protocols to maintain communication in dynamic and decentralized environments.
The Ad-hoc On-Demand Distance Vector (AODV), Dynamic Source Routing (DSR), and Dynamic MANET On-demand (DYMO) are three widely used protocols, each with distinct strategies for route discovery and maintenance. This paper analyze their performance using metrics such as Packet Delivery Ratio (PDR), end-to-end delay, throughput, and routing overhead. Simulations conducted with OMNeT++ explore the impact of network disruptions, node failures, and physical obstacles on protocol efficiency. The results offer a comparative analysis, highlighting the strengths and limitations of each protocol across different scenarios. These insights provide useful guidance for selecting suitable routing protocols for mobile applications and limited-resource environments.
|
| Location |
Informatics Institute
Room 317
|
|
| 6 |
| Title |
Survey of Automated Theorem Provers:
Classification, Comparison, and Application in Formal Verification |
| Speaker |
Lina Ben Haj Yahya |
| Date |
December 25th, 2024 |
| Time |
14:00 (GMT +3) |
| Advisor |
Mehmet Tahir Sandıkkaya |
Abstract
Theorem provers are essential tools in formal verification, ensuring the correctness of complex systems. This paper will survey and classify prominent automated and interactive theorem provers (ATPs and ITPs), including AVISPA, Tamarin, Coq, Agda, Maude, HOL, Lean, and others. It explores their logical foundations, such as first-order and higher-order logics, SMT, and QBF, and examines their application in constraint programming, mixed-integer programming (MIP), and dependent type theory. The study compares these tools based on performance, scalability, and ease of use, providing insights for their use in formal verification scenarios and identifying opportunities for future advancements.
|
| Location |
Informatics Institute
Room 317
|
|
| 7 |
| Title |
Analysis of the Success of Password Reset Methods in Terms of Security and Usability |
| Speaker |
Ahmet Furkan İnal |
| Date |
December 25th, 2024 |
| Time |
14:30 (GMT +3) |
| Advisor |
Kemal Bıçakçı |
Abstract
A key element of digital security is password reset techniques, which strike a balance between user ease and the requirement for robust protection. This study looks at popular password reset methods from a security and usability standpoint, such as two-factor authentication, email verification, security questions, and SMS verification. This study illustrates the trade-offs between more user-friendly but vulnerable approaches and more safe but complex ones by examining both security flaws and the user experience.
The study evaluates the effectiveness of these techniques in practical applications by combining user surveys and a review of the literature. The results show that although simpler approaches increase user satisfaction but could compromise security, security-focused approaches improve protection at the expense of usability. In order to guarantee accessibility and safety, the study ends with suggestions for developing more secure and user-friendly password reset procedures.
|
| Location |
Informatics Institute
Room 317
|
|
| 8 |
| Title |
Efficient Reporting of Data Based on Stochastic Time Frames |
| Speaker |
Emre Altunel |
| Date |
December 25th, 2024 |
| Time |
15:00 (GMT +3) |
| Advisor |
Murat Okatan |
Abstract
In many fields, data is governed by stochastic processes, where event timings are influenced by probabilistic patterns rather than deterministic rules. A significant challenge in such datasets is the difficulty in efficiently capturing time intervals, particularly when events are ordered by their start times, but their end times do not adhere to predictable patterns and vice versa. This paper addresses this issue by proposing methods for transforming stochastic time frame data to facilitate its efficient reporting and analysis. We introduce a novel method that integrates probabilistic modeling with adaptive techniques to optimize the capture of irregular time gaps, enabling more effective data reporting and interpretation.
|
| Location |
Informatics Institute
Room 317
|
|
| 9 |
| Title |
Detection of Cancerous Areas In Prostate Pathological Images Using Deep Learning |
| Speaker |
Furkan Yaşar |
| Date |
January 8th, 2025 |
| Time |
13:30 (GMT +3) |
| Advisor |
Volkan Müjdat Tiryaki |
Abstract
Prostate cancer detection in pathological images is a critical task in medical diagnostics, often requiring expert analysis for accurate identification. Traditional methods for detecting cancerous areas in histopathological images are time-consuming and dependent on the skill of pathologists. Recent advancements in deep learning have demonstrated significant potential in automating and improving the accuracy of cancer detection. This study aims to explore the application of deep learning techniques for detecting prostate cancer regions in pathological images. We plan to propose a convolutional neural network (CNN)-based approach that will leverage a large dataset of prostate biopsy images to identify and localize cancerous areas.
The model will be trained and evaluated using a combination of image augmentation and transfer learning to enhance its performance. It is expected that the results will demonstrate the effectiveness of deep learning in accurately identifying cancerous regions, with performance metrics such as sensitivity, specificity, and area under the curve (AUC) showing substantial improvement over traditional methods. This work will highlight the potential of deep learning models to assist pathologists in diagnosing prostate cancer, offering a promising avenue for early detection and treatment.
|
| Location |
Informatics Institute
Room 317
|
|
| 10 |
| Title |
Machine-Learning for Optimal Detection of Inflammatory Cells in the Kidney |
| Speaker |
Can Görkem Güneş |
| Date |
January 8th, 2025 |
| Time |
14:00 (GMT +3) |
| Advisor |
Volkan Müjdat Tiryaki |
Abstract
This project aims to automate the detection of inflammatory cells in kidney transplant biopsies. The aim is to develop a machine learning model that detects inflammatory cells such as lymphocytes and monocytes, especially in PAS-stained biopsy samples. Current manual methods are both time-consuming and prone to error, so the model to be developed aims to increase accuracy and speed up this process. The performance of the model will be evaluated by testing it on real data sets, and a tool with potential use in the field of pathology will emerge.
|
| Location |
Informatics Institute
Room 317
|
|
| 11 |
| Title |
Human Detection with Deep Learning from Images taken by Unmanned Aircraft |
| Speaker |
Ali Karakuş |
| Date |
January 8th, 2025 |
| Time |
14:30 (GMT +3) |
| Advisor |
Yusuf Sinan Akgül |
Abstract
Human detection from images captured by unmanned aircraft systems (UAS) has become a critical task in applications such as search and rescue, surveillance, and disaster management. This paper presents a deep learning-based approach for detecting humans in aerial imagery, addressing the unique challenges posed by varying altitudes, angles, and environmental conditions in such images. We leverage convolutional neural networks (CNNs) to extract robust features from aerial images, enabling the detection of humans with high accuracy. Our model is trained on a custom dataset of UAS-captured images, incorporating data augmentation techniques to enhance generalization to diverse scenarios. We evaluate our method on both publicly available and real-world datasets, demonstrating its ability to detect humans in complex environments with varying resolutions and occlusions. Experimental results show that the proposed deep learning approach significantly outperforms traditional methods in terms of precision, recall, and overall detection performance, making it well-suited for real-time deployment in UAS-based applications.
|
| Location |
Informatics Institute
Room 317
|
|