Master Degree / Yüksek Lisans Tezleri

Permanent URI for this collectionhttps://hdl.handle.net/11147/3008

Browse

Search Results

Now showing 1 - 10 of 152

Estrus Detection in Cows With Deep Learning Techniques
(01. Izmir Institute of Technology, 2024) Arıkan, İbrahim; Ayav, Tolga; Soygazi, Fatih
Accurately predicting the estrus period is essential for enhancing the efficiency and lowering the costs of artificial insemination in livestock, a crucial sector for global food production. Precisely identifying the estrus period is critical to avoid economic losses such as decreased milk production, delayed calf births, and loss of eligibility for government subsidies. Since the most obvious movement that needs to be detected during the fertilization period is mounting, it is important to detect this movement. Since manual detection of this movement is difficult and costly, automated methods were needed. Therefore, it is thought that deep learning-based methods can be applied to detect the mounting moment. The proposed method detects the estrus period using deep learning and XAI (Explainable Artificial Intelligence) techniques. Deep learning-based mounting detection is performed using CNN, ResNet, VGG-19 and YOLO-v5 models. The ResNet model in this proposed study detects mounting movement with 99% accuracy. Explainability of deep learning models describes features that aid in decision-making in detecting mounting motion. Grad-CAM and Gradient Inputs models, which are XAI techniques, are used for the black box behind the proposed models. The developed deep learning models reveal that they focus on the udder and back area of the cows during the decision-making phase. In addition, how successfully the Grad-CAM and Gradient Inputs models, which are the XAI models used for the explainability of the deep learning models trained in this study, performed the explanation process was measured by calculating the 'faithfulness', 'maximum sensitivity' and 'complexity' metrics.
Combining Persona and Argument in Dialogue
(2024) Güzel, Şükrü; Tekir, Selma
The increasing popularity of personalized dialogue systems has gained momentum as people's desire for human-like interaction grows. This thesis aims to increase persona-consistent responses in personalized dialogue systems. A data augmentation method was used to enhance the persona consistency of dialogue systems. This technique utilized Large Language Models' few-shot learning capabilities to add counterfactual sentences to the dialogue. GPT 3.5 and Llama 2 models were used to generate counterfactual sentences using the few-shot prompting method. The augmentation method was applied to every dialogue in the PersonaChat dataset that did not originally contain a counterfactual sentence. Evaluation using the state-of-the-art personalized dialogue generation study showed that the persona-consistency results of the dataset augmented with the GPT 3.5 model showed better performance when assessed using metrics.
Predicting Software Size From Requirements Written in Natural Language: a Generative Ai Approach
(01. Izmir Institute of Technology, 2024) Kennouche, Dhıa Eddıne; Demirörs, Onur
In project management, software size measurement represents a critical process aimed at visualizing a project. This quantification is pursued independently of the specific technologies or technical decisions adopted during the project's development phase. Among the various methodologies employed for this purpose, the COSMIC Functional Size Measurement (FSM) and Event Points are used to facilitate such assessments. These methodologies are instrumental in offering a standardized approach for measuring software size, yet they inherently demand a considerable amount of manual effort. Furthermore, these methods require the manual extraction of Objects of Interest and Event Names, adding to the labor-intensive nature of the process. In response to these challenges, this thesis implements a suite of Artificial Intelligence (AI)-based methods that have dramatically transformed the measurement process. These innovative approaches encompass the creation of a Regression Model that predicts software sizes with remarkable accuracy, a Summarization Model that automates the extraction of Event Names, and a finely tuned Large Language Model (LLM) that generates Objects of Interest with a significant precision. The adoption of these AI-driven techniques has proven to be highly successful, substantially minimizing the manual effort traditionally required in software size measurement and thereby greatly enhancing both efficiency and reliability of estimation practices. Together, these AI-based methodologies represent a significant advancement in software size measurements, offering a more streamlined and efficient approach. By reducing the reliance on manual processes, these methods not only enhance the accuracy and reliability of measurements but also contribute to a more agile project management environment.
Modeling Microservice Based Applications: Model Lives Inside Code Approach
(01. Izmir Institute of Technology, 2024) Ersoy, Eyüp Fatih; Demirörs, Onur
In today's software development, maintaining consistent documentation is crucial for sharing and preserving team knowledge. As projects grow more complex, developers need to quickly understand and maintain code. However, keeping documentation aligned with business logic without unnecessary technical details is challenging. Traditional visualization tools like UML, sequence, and activity diagrams focus on object-oriented approaches and often require manual updates, making them less suitable for event-based systems like microservices. To address these issues, the tool Docupyt was developed using eEPC (Extended Event Process Chains) as the main modeling approach. Docupyt is designed with three key principles: ease of use, simplicity (including only necessary logic), and reactivity (representing event-based systems). eEPC notation helps analyze problems and represent changing logic during development, accommodating fast-changing requirements. It supports both high and low-level process definitions and focuses on business logic without extraneous technical details. Generated directly from code through simple commenting, this approach simplifies updating documentation as the code changes, reducing maintenance costs. Using the design science research method, Docupyt was validated in a case study, demonstrating it is user-friendly and provides adequate detail without being overly technical. Its main advantage is keeping documentation in sync with code logic, easing updates.
Learning Citation-Aware Representations for Scientific Papers
(01. Izmir Institute of Technology, 2024) Çelik, Ege Yiğit; Tekir, Selma
In the field of Natural Language Processing (NLP), the tasks of understanding and generating scientific documents are highly challenging and have been extensively studied. Comprehending scientific papers can facilitate the generation of their contents. Similarly, understanding the relationships between scientific papers and their citations can be instrumental in generating and predicting citations within the text of scientific works. Moreover, language models equipped with citation-aware representations can be particularly robust for downstream tasks involving scientific literature. This thesis aims to enhance the accuracy of citation predictions within scientific texts. To achieve this, we hide citations within the context of scientific papers using mask tokens and subsequently pre-train the RoBERTa-base language model to predict citations for these masked tokens. We ensure that each citation is treated as a single token to be predicted by the mask-filling language model. Consequently, our models function as language models with citation-aware representations. Furthermore, we propose two alternative techniques for our approach. Our base technique predicts citations using only the contexts from scientific papers, while our global technique incorporates the titles and abstracts of papers alongside the contexts to improve performance. Experimental results demonstrate that our models significantly surpass the state-of-the-art results on two out of four benchmark datasets. However, for the remaining two datasets, our models yield suboptimal results, indicating potential for further improvement. Additionally, we conducted experiments on sampled datasets to examine the effects of inherent factors on the datasets and to identify correlations between these factors and our results.
Privacy-Preserving Rare Disease Analysis With Fully Homomorphic Encryption
(01. Izmir Institute of Technology, 2023) Akkaya, Güliz; Erdoğmuş, Nesli; Akgün, Mete
Rare diseases severely affect many people across the world at the present time. Researchers conduct studies to understand the reasons behind rare diseases and as a result of this research, diagnosis, and treatment methods are developed. Rare disease analysis is performed to specify the disease-causing variants on the genome data of patients. The researchers need access to as much genome data as possible to find causing variants of rare diseases. On the other hand, the genome data of patients should be protected because it can be used to detect the identity of individuals. The researchers are not able to share the genome data of patients easily because of regulations such as General Data Protection Regulation (GDPR). For this reason, rare disease analysis should be performed in a secure way that protects the privacy of patients while enabling the collaboration of multiple medical institutions. In this context, a privacy-preserving collaborative system for rare disease analysis should be provided. This thesis study focuses on the utilization of fully homomorphic encryption, a method that enables unlimited number of operations to be performed on encrypted data, for privacy-preserving collaborative rare disease analysis. Two different methods, the boolean circuit method, and the integer arithmetic method, are implemented to perform rare disease analysis on the encrypted genome data to find disease-causing variants, and various experiments are performed to assess the efficiency of the proposed methods.
Multi-Frame Super-Resolution Without Priors
(01. Izmir Institute of Technology, 2023) Gülmez, Veli; Özuysal, Mustafa
There are mainly two types of super-resolution methods: traditional methods and deep learning methods. While traditional methods define closed-form expressions with assumptions, deep learning methods rely on priors learned from data sets. However, both of them have disadvantages such as being too simple and having strong trust in priors. We focus on how to generate a high-resolution image using low-resolution images without priors by utilizing spatial hash encoding. We propose a grid-based super-resolution model using spatial hash encoding to map coordinate information into higher dimensional space. Our aim is to eliminate long training times and not rely on priors from data sets that are not able to cover all real-world scenarios. Therefore, our proposed model is able to do task- specific super-resolution without priors and eliminate potential hallucination effects caused by wrong priors.
Enrichment of Turkish Question Answering Systems Using Knowledge Graphs
(01. Izmir Institute of Technology, 2023) Çiftçi, Okan; Tekir, Selma; Soygazi, Fatih
In the era of digital communication, the ability to effectively process and interpret human language has become a key research area. Natural Language Processing (NLP) has emerged as a field that enables machines to better understand and analyze human language. One of the most important applications of NLP is the development of question answering systems, which are essential in various domains such as customer service, search engines, and chatbots. To answer incoming queries, question answering systems rely on knowledge graphs as a reliable source. This thesis proposes a Turkish Question Answering (TRQA) system that utilizes a knowledge graph. The research focuses on the automatic construction of a knowledge graph specific to the film industry, as well as the creation of a multi-hop question-answering dataset that can be queried from this graph. Building upon these constructions, we develop a deep learning based method for answering questions using the constructed knowledge graph. The constructed knowledge graph is compared with various knowledge graphs presented in the literature using DistMult, ComplEx and SimplE methods for the link prediction task. Additionally, the proposed question answering system is compared with the baseline study and compared with a generative large language model through quantitative and qualitative analyses.
Reproducibility Assessment of Research Code Repositories
(01. Izmir Institute of Technology, 2023) Akdeniz, Eyüp Kaan; Tekir, Selma
The growth in machine learning research has not been accompanied by a corresponding improvement in the reproducibility of the results. This thesis presents a novel, fully-automated end-to-end system that evaluates the reproducibility of machine learning studies based on the content of the associated GitHub project's Readme file. This evaluation relies on a readme template derived from an analysis of popular repositories. The template suggests a structure that promotes reproducibility. Our system generates a reproducibility score for each Readme file assessed, and it employs two distinct models, one based on section classification and the other on hierarchical transformers. The experimental outcomes indicate that the system based on section similarity outperforms the hierarchical transformer model. Furthermore, it has a superior edge concerning explainability, as it allows for a direct correlation of the scores with the respective sections of the Readme files. The proposed framework provides an important tool for improving the quality of code sharing and ultimately helps to increase reproducibility in machine learning research.
Row Following and Altitude Estimation With Uav Images for Agricultural Fields
(01. Izmir Institute of Technology, 2023) Yörük, Burak; Baştanlar, Yalın
Traditional methods in agriculture involve the use of tractors; however, more than 10\% of the planted fields suffer from harvest losses due to these vehicles. Moreover, tractors cannot enter all agricultural lands, thus reducing the available field for planting. After heavy rainfall, mud and other effects prevent these vehicles from accessing arable field, and processes such as crop spraying take significantly longer. In the past, aerial spraying methods using high altitude aircraft were attempted to overcome these problems; however, this method was banned in many areas due to the insufficient altitude and the harmful effects of chemical dispersion outside the fields. Nowadays, UAVs present a better alternative, and aerial spraying methods are regaining popularity. However, these vehicles can still cause errors when flying with a human operator, and their flight times are limited due to inadequate battery capacity. Therefore, the development of UAVs capable of autonomous flight reduces operator costs. However, during flight, liquid changes in the pesticide tanks hinder the UAV's ability to spray pesticides autonomously at a fixed altitude and prevent unwanted pesticide dispersion in undesirable rows. The thesis study provides following of plant rows on UAV images and making altitude estimation from camera images. In this way, it ensures that the UAVs in agricultural areas can stay at a fixed altitude for appropriate spraying and irrigation and prevents the spread of pesticides to unwanted rows.

Master Degree / Yüksek Lisans Tezleri

Browse

Filters

Settings

Sort By

Results per page

Search Results