Repository logoGCRIS

Scopus İndeksli Yayınlar Koleksiyonu

Permanent URI for this collectionhttps://hdl.handle.net/20.500.12573/395

Browse

Filters

2024 - 20265

Settings

Subject: search.filters.subject.Federated Learning

Search Results

Now showing 1 - 5 of 5
  • Article
    Threat Landscape of Edge-AI-Assisted Connected Autonomous Vehicles (CAV)
    (Elsevier B.V., 2026) He, Ligang; Maple, Carsten; Atmaca, Ugur Ilker; Kasyap, Harsh; Nezhad, Mahshid Mehr; Atmacaa, Ugur Ilker; Nezhada, Mahshid Mehr; Kasyapa, Harsh
    Connected Autonomous Vehicles (CAVs) powered by Edge Artificial Intelligence (Edge-AI) are revolutionising intelligent transportation systems through real-time data processing for obstacle avoidance, adaptive learning of dynamic traffic patterns, and improved decision-making in safety-critical tasks like traffic sign recognition. However, deploying machine learning (ML) models directly at the vehicle level introduces new security vulnerabilities, making CAVs susceptible to adversarial attacks that can compromise system integrity, reliability, and safety. These attacks primarily target onboard ML systems, highlighting the need for a structured threat modelling approach to identify and mitigate risks. This paper investigates the threat landscape of Edge-AI-assisted CAVs, focusing on the CAV layer as a critical attack surface within the hierarchical system architecture. Employing the STRIDE framework, we analyse vulnerabilities across four stages of the ML lifecycle: input data, model training, aggregation and inference. We present adversarial concept drift as a case study of an evolving threat in which malicious actors introduce subtle data manipulations over time to degrade ML model performance while avoiding detection. Through attack tree analysis and experimental evaluation, we show how adversarial concept drift propagates through the system, ultimately affecting its reliability. To the best of our knowledge, this is the first study to illustrate the impact of gradual adversarial concept drift on robustness. Our results reveal that gradual drifts are harder to detect than sudden ones, even when using Byzantine-robust defenses. These findings underscore the limitations of existing security mechanisms and emphasise the urgent need for adaptive, resilient countermeasures to address the dynamic, evolving threat landscape facing Edge-AI-enabled CAV systems.
  • Article
    Citation - Scopus: 1
    eTNT: Enhanced Textnettopics With Filtered LDA Topics and Sequential Forward / Backward Topic Scoring Approaches
    (Science and Information Organization, 2024) Voskergian, Daniel; Jayousi, Rashid; Bakir-Güngör, Burcu
    TextNetTopics is a novel text classification-based topic modelling approach that focuses on topic selection rather than individual word selection to train a machine learning algorithm. However, one key limitation of TextNetTopics is its scoring component, which evaluates each topic in isolation and ranks them accordingly, ignoring the potential relationships between topics. In addition, the chosen topics may contain redundant or irrelevant features, potentially increasing the feature set size and introducing noise that can degrade the overall model performance. To address these limitations and improve the classification performance, this study introduces an enhancement to TextNetTopics. eTNT integrates two novel scoring approaches: Sequential Forward Topic Scoring (SFTS) and Sequential Backward Topic Scoring (SBTS), which consider topic interactions by assessing sets of topics simultaneously. Moreover, it incorporates a filtering component that aims to enhance topics' quality and discriminative power by removing non-informative features from each topic using Random Forest feature importance values. These integrations aim to streamline the topic selection process and enhance classifier efficiency for text classification. The results obtained from the WOS-5736, LitCovid, and MultiLabel datasets provide valuable insights into the superior effectiveness of eTNT compared to its counterpart, TextNetTopics. © 2024 Elsevier B.V., All rights reserved.
  • Conference Object
    Citation - Scopus: 1
    Words Speak Louder Than Actions: Decoding Emotions Through NLP
    (Institute of Electrical and Electronics Engineers Inc., 2024-10-26) Paksoy, Melda; Bakal, Gokhan
    Emotion detection in text remains a significant challenge in Natural Language Processing due to human emotions' complexity and subtle nuances. This paper presents multiple experimental models for emotion classification using an up-to-date dataset curated to address 13 emotions implied in Twitter posts. We evaluated various machine learning (ML) models, including Logistic Regression, Random Forest, SVM, and XGBoost, alongside deep learning (DL) architectures such as LSTM and CNN. Our results demonstrate the efficacy of deep learning models, particularly the CNN model by achieving an impressive F1 score of 0.99. This study contributes to emotion detection capabilities, paving the way for more nuanced and accurate sentiment analysis (SA) in various text analysis applications. © 2025 Elsevier B.V., All rights reserved.
  • Conference Object
    Citation - Scopus: 2
    Many-to Transfer Learning on Motor Imagery BCI
    (Institute of Electrical and Electronics Engineers Inc., 2024-12-11) Altindis, Fatih; Yilmaz, Bulent; Congedo, Marco
    This paper presents many-to-many domain adaptation strategy, named group learning, for motor imagery brain-computer interfaces (BCIs). Group learning, grounded in Riemannian geometry, simultaneously aligns multiple domains in a unified model, whereas fast alignment approach integrates new, unseen domains without re-estimating alignment matrices for all domains. Group learning creates a single machine learning model using data from previous subjects and/or sessions. Fast alignment utilizes the already trained model for an unseen domain without requiring any additional classifier training. The tests on five publicly available motor imagery databases demonstrate the robustness of group learning against negative learning. The classification accuracy scores of binary and multiclass databases show comparable, if not superior, performance to conventional subject-wise learning method. © 2025 Elsevier B.V., All rights reserved.
  • Conference Object
    Citation - Scopus: 1
    A Federated Learning Framework for Classifying the Images in Ultrasonic Nondestructive Testing
    (Institute of Electrical and Electronics Engineers Inc., 2024-10-26) Gulsen, Abdulkadir; Hacilar, Hilal; Kolukisa, Burak; Bakir-Güngör, Burcu
    Ultrasonic inspection is a critical technique in non-destructive testing that ensures the safety and integrity of the material by detecting internal defects. Defect classification within this context is vital for preventing failures and extending the lifespan of materials. However, the advancement of ultrasonic testing technology is hindered by a scarcity of publicly available, realistic datasets, which are essential for developing accurate models. To address these challenges, this paper introduces a Federated Learning (FL) framework employing a Convolutional Neural Network (CNN) model for defect classification using ultrasonic inspection images. This innovative approach allows for the decentralized training of models on private datasets without the need for data exchange, thus preserving data privacy. Our comparative analysis demonstrates that the FL achieves performance comparable to traditional methods while maintaining the confidentiality of sensitive information. The framework also proves to be robust and scalable with an increase in the number of participating clients. This pioneering study highlights the potential of FL in transforming ultrasonic defect classification and suggests possibilities for its application in other areas of non-destructive testing where publicly available datasets are scarce. These findings would encourage researchers to develop a federated platform for enhanced collaboration and explore advanced CNN architectures to improve training efficiency. © 2025 Elsevier B.V., All rights reserved.