https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4117 Thu, 26 Mar 2026 11:08:41 GMT 2026-03-26T11:08:41Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4946 Title: Investigating the Effect of Different Subflow MTUs on MPTCP Throughput Authors: Manoratne, C.S. Abstract: Abstract Multipath TCP (MPTCP) is an extension of the traditional Transmission Control Protocol (TCP) that enables simultaneous use of multiple network paths between two endpoints, offering improved resilience, throughput, and resource utilization. While existing MPTCP implementations perform well in homogeneous environments, their behavior in heterogeneous networks, particularly when subflows have different Path Maximum Transmission Units (MTUs) remains underexplored. This research investigates the impact of varying subflow MTUs on MPTCP throughput using a controlled emulation environment. Experimental analysis reveals that the default Linux MPTCP implementation fails to utilize subflows with smaller MTUs effectively, primarily due to limitations in the Path MTU Discovery (PMTUD) mechanism and a unified Maximum Segment Size (MSS) approach. Kernel-level modifications were introduced to enable MTU probing on a per-subflow basis, allowing dynamic MSS adjustment and improved subflow utilization. Results demonstrate a notable improvement in throughput and path diversity post-modification, highlighting the importance of MTU-aware scheduling and adaptive probing techniques. The findings suggest potential directions for enhancing MPTCP performance in heterogeneous environments and contribute to the ongoing development of more robust multipath communication protocols. Fri, 25 Apr 2025 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4946 2025-04-25T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4945 Title: Applicability of Transfer Learning on Sinhala Named-Entity Recognition Authors: Abeynayaka, A.G.K.C Abstract: ABSTRACT Named Entity Recognition (NER) is a preliminary task in Natural Language Processing. NER has evolved from relying on rule-based mechanisms to utilizing neural networks. NER is a pretty much resolved matter in the English language. The Sinhala language faces the issue of data scarcity due to its complexities with dataset extraction. Manual annotation of a Sinhala-labeled dataset is a laborious task. Entity recognition solely depends on a tagged dataset in a specific language, but due to data limitations, it’s hard to do experiments on NER models in Sinhala. However, most of low-resource NLP researches shows remarkable improvement with the knowledge transferring mechanism, which is known as transfer learning. This research suggests a Sinhala NER model based on transfer learning, considering monolingual and multilingual approaches. An Indic language model is fine-tuned for the target Sinhala NER model during both approaches. The IndicBERT(Kakwani et al. 2020) model is chosen as the source model due to its similarity with Sinhala. The evaluations were done on monolingual and multilingual datasets. For the monolingual dataset, a separate dataset was created using a weakly supervised automatic method that contains six different categories. The multilingual dataset was created with a Bengali dataset. The final transfer learning model was trained on hyperparameter tuning followed by an augmented dataset from monolingual data. It showed a moderate precision of 48.21%. The baseline CRF model showed a macro precision of 90% and a macro F1-score of 61% showing that CRF is applicable in normal contexts. Mon, 28 Apr 2025 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4945 2025-04-28T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4944 Title: Real-Time 3D Avatar Modeling for AR using Human Pose and Actions in Resource-Constrained Web Environments Authors: Galappaththy, S.R. Abstract: Abstract Real-time 3D avatar animation using standard webcams offers immense potential for immersive communication and interaction within web-based Augmented Reality (AR). However, existing solutions often struggle with accessibility and generalisation, either requiring specialised hardware or relying on a single, high-fidelity pose estimation input that may be too resource-intensive for many web environments. This creates a significant gap, as these systems cannot easily adapt to the diverse quality and format of data from various readily available pose estimators or gracefully handle common real-world challenges like partial user visibility. This research addresses these limitations by employing a Design Science Research methodology to design, implement, and evaluate a novel, generalized middleware pipeline for real-time, webcam-based 3D avatar animation, operating entirely within standard web browsers. The core contribution is a modular JavaScript-based architecture centered around a biomechanicallyaware canonical pose representation aligned with the VRM humanoid standard. The methodology involves developing an adaptive input adapter for heterogeneous data (from MoveNet, BlazePose, YOLO-Pose, etc.), a pose processor for heuristic 2D-to-3D lifting and robust inference of occluded joints using data-driven priors from H36M, and a flexible retargeting module. Experimental results demonstrate the pipeline’s ability to successfully process diverse inputs and drive plausible, full-body avatar animations in real-time. Notably, the system generates coherent motion even from sparse 2D keypoint data where simpler direct mapping would fail. Performance analysis indicates viability on desktop/laptop browsers and feasibility on mobile devices with lighter-weight estimators. This research presents a significant step towards more accessible and flexible real-time avatar systems for the web platform, providing a practical and extensible foundation for future advancements in web-based embodied interaction. Mon, 30 Jun 2025 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4944 2025-06-30T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4943 Title: Investigating Linux Random Number Generator for Virtualization Detection from the Non Privileged User Space Authors: Harshani, A. A. D. Abstract: Abstract The detection of virtualization presence is a critical problem in malware analysis, where malicious software may attempt to identify if it is being tested within a virtual environment. Existing methods often require special privileges, creating a gap for non privileged approaches. This study investigates the feasibility of detecting virtualization presence from the non privileged user space by analyzing the behaviour of the Linux Random Number Generator (LRNG), a component not previously used for this purpose. A series of experiments were conducted across bare metal and virtual environments under varying impact levels to assess differences in random number generation rates and quality. The evaluation included both single and multiple VM setups, across desktop, private, and public cloud infrastructures. Results revealed measurable distinctions in LRNG behaviour between bare metal and virtual environments through distinct timing distributions, where early peaks were observed. These early peaks refer to instances where random number generation took significantly longer in virtual environments in the beginning compared to bare metal systems. Additionally, differences in dispersion patterns across bare metal and virtual environments were identified, which were collectively used for the detection of virtualization environments through derived thresholds, achieving a detection accuracy of up to 94.44%. The study also examined the role of entropy enhancing tools designed to improve the randomness of generated data, in obscuring virtualization presence. The results proved approach is ineffective, suggesting the need for further research into obscure such detection. The influence of the operating system on LRNG behaviour was identified as a significant factor, with notable differences observed between Debian and Red Hat based Linux systems. These findings demonstrate the potential of LRNG characteristics for non privileged virtualization detection with a novel direction. Unlike traditional detection methods that rely on privileged access, this approach operates entirely from the user space, demonstrating the feasibility of using user space behaviours to address virtualization detection challenges and opening possibilities for further research in this domain. Future work should focus on extending the scope of these findings, addressing the limitations identified, and exploring additional methods to enhance the robustness of detection and obfuscation techniques. Thu, 26 Jun 2025 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4943 2025-06-26T00:00:00Z