Thesis (Master of Science/Engineering in Computer Science and Engineering)

Permanent URI for this collectionhttps://hdl.handle.net/10361/28432

Browse

Now showing 1 - 20 of 105

Open Access
Real-time face recognition system leveraging facenet and VGGFace fusion embedding, and siamese network with ArcFace and triplet loss function
(BRAC University, 2026-01) Haque, Md. Ashfaqul; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
The face recognition system can be defined as the identification of the unique biometric facial features of an individual through image or video frames. For CCTV-based surveillance the facial recognition system automates the identification of any person applied on live surveillance stream. But in CCTV surveillance system the facial recognition endures unique challenges like low resolutions, extreme poses, illumination variations, etc. There, a long span of development has been seen in this system from the geometric landmark distancing to deep-learning approaches, but a significant gap between the real-life surveillance and controlled benchmark conditions is observed. This thesis presents a comprehensive study on the development of a face recognition system designed for use with CCTV footage, with a focus on addressing challenges such as side-profile recognition and low-resolution images. Traditional face recognition models often perform poorly in real-world surveillance scenarios due to the quality of footage, low-resolution face crops, illumination variations, the varied angles of faces captured, different head orientations, occlusion causing partial face visibility, and motion effects. To overcome these limitations, this research explores the integration of a Siamese network with FaceNet and VGGFace, creating a hybrid model that leverages the unique strengths of each component. The Siamese network is employed to compare facial data, while VGGFace and FaceNet generate vigorous embeddings for accurate feature extraction, and a hybrid of Triplet Loss and ArcFace respectively enhances discriminative capabilities and combining through directing better embedding generations for similar faces, ensuring precise identification across challenging inputs. A custom dataset was constructed to simulate real-world CCTV environments, incorporating diverse facial angles and resolutions to rigorously test the model’s performance. The primary aim of the research is to evaluate the system’s ability to accurately recognize faces under non-ideal conditions, optimize computational efficiency, and assess the scalability of the model for large-scale surveillance systems. Through extensive experimentation and iterative refinement, the proposed hybrid model demonstrates improved recognition accuracy in low-quality, a variety of angled, and different expression face images, which contributes to advancements in the field of computer vision and offers practical implications for security and surveillance applications. As a dataset of 5409 face crop images from 86 individual person have been comprised here and later balanced to a total of 4300 samples and the proposed hybrid model has achieved an accuracy of 91.01% with 91.16% f1-score and ROC-AUC 0.9665, leading performance over the deepface models in comparison. Also the model have shown a competitive performance on the benchmark datasets of LFW, CelebA and VGGFace2, achieving an accuracy of 99.28%, 98.57% and 96.82% with f1-score of 98.71%, 98.46% and 96.27% respectively. Though the gap of benchmark datasets and CCTV footage data, it demonstrates a practical applicability for using in face recognition operations.
Open Access
Behind the firewall: A study on structured vulnerability assessment of the websites of the tertiary educational institutes of Bangladesh focusing on intuitive attack vectors to realize the cybersecurity practices in web design of academia
(BRAC University, 2026-03) Khan, Rafia Tabassum; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
The rapid expansion of web-based services in higher education has significantly increased the exposure of university websites to cyber threats. In Bangladesh, many academic institutions rely on publicly accessible web applications for administrative, academic, and informational purposes, making them potential targets for website infections and exploitation. This research examines the security posture of selected public and private university websites in Bangladesh by identifying, analyzing, and comparing prevalent web application vulnerabilities. The study adopts a structured vulnerability assessment methodology grounded in the OWASP framework, focusing on common attack vectors such as SQL Injection (SQLi), Cross-Site Scripting (XSS), authentication weaknesses, insecure configurations, and improper input validation. Controlled and non-destructive testing techniques were employed using industry-standard tools in a safe and ethical virtual environment, Kali Linux. Data collected from the assessments were systematically analyzed using statistical methods to evaluate vulnerability frequency, severity, and distribution across institutional categories. The results reveal discernible differences between public and private universities, with variations in defensive practices, exposure levels, and maintenance strategies. Several recurring vulnerabilities indicate gaps in secure coding practices, patch management, and security awareness. The findings highlight the need for inclusive and proactive cybersecurity strategies, regular vulnerability assessments, and alignment with established web security standards. This research contributes to the understanding of website infection risks within academic institutions in developing regions and provides practical insights and recommendations to strengthen web application security, improve resilience against cyberattacks, and promote responsible cybersecurity practices in higher education.
Open Access
Learning to win: A process reward model for competitive machine learning agents
(BRAC University, 2026-03) Ahmed, Sajjad; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Large Language Model (LLM)-based coding agents are increasingly deployed for multi-step data science tasks, yet no systematic study has examined how these agents behave across diverse competitive problems. This thesis presents a two-part investigation using 194 human-supervised agent human feedback JSONs across 103 Kaggle competitions (1,584 turns total). First, we conduct a comprehensive empirical study. We also propose and show that process-oriented quality metrics. Second, we develop a Process Reward Model (PRM) for data science workflows—the first application of PRMs beyond mathematical reasoning—that predicts whether an intermediate step will improve competition scores. Using 49 tabular features across 6 feature groups, we train XGBoost and LightGBM baselines and conduct a full feature ablation study. Code features (code length, imports, deltas) emerge as the strongest individual tabular predictors. We further fine-tune Qwen2.5-Coder-3B with QLoRA as a hybrid language-model-based PRM (v2), incorporating code snippets and tabular features into semantic prompts, which achieves AUROC 0.656—surpassing the best tabular baseline (0.530) by 12.6 points—by leveraging semantic content from agent plans, reflections, and code. To establish the necessity of domain-specific finetuning, we evaluate zero-shot frontier models (i.e., GPT-5, Gemini 3.0 Flash/Pro, Llama 3.3 70B) on the same task. GPT-4o achieves the highest zero-shot AUROC (0.709), outperforming our fine-tuned 3B model, while most other frontier models fall to majority class prediction (AUROC 0.500). This demonstrates that while powerful zero-shot models can reason about the task, domain-specific fine-tuning remains valuable for smaller models, achieving competitive performance with 200x fewer parameters. Ultimately, our research demonstrates that observable code signals are significantly more reliable than an agent’s articulated plans for predicting success, providing a robust foundation for the development of real-time guidance systems in competitive data science.
Open Access
Knowledge distillation in split learning for heterogeneous clinical tabular data
(BRAC University, 2026-03) Farheen, Tasneem Jahan; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
The growing availability of clinical data across institutions creates new opportunities for improved disease risk prediction. However, privacy regulations, institutional policies, and limited labeled data restrict centralized model training, particularly in low-resource settings. To address these challenges, this study presents a privacypreserving framework that integrates split learning and knowledge distillation for diabetes risk prediction using heterogeneous tabular datasets. This work systematically evaluates two complementary transfer mechanisms. The first approach utilizes logit-level knowledge distillation to transfer predictive decision boundaries from high-capacity teacher models, collaboratively trained on large population datasets, to lightweight student models in low-resource target domains. The second approach implements encoder-level distillation to align latent feature representations within a split learning architecture. Both approaches maintain strict data privacy by keeping raw patient data local and exchanging only intermediate activations. Experimental results under few-shot supervision indicate that logit-level distillation substantially improves discriminative performance, particularly in extremely low-label scenarios while stabilizing training across heterogeneous domains. Encoder-level distillation further improves probabilistic calibration and representation alignment under crossdomain distribution shifts. These findings underscore the significance of structured knowledge transfer in privacy-preserving clinical machine learning and offer practical recommendations for deploying reliable models in heterogeneous healthcare environments.
Open Access
Explainable dual-teacher knowledge distillation with confidence-aware knowledge filtering for lightweight wound segmentation
(BRAC University, 2026-04) Dofadar, Dibyo Fabian; Galib, Syed Md.; Department of Computer Science and Engineering
Medical image segmentation plays a crucial role in automated wound assessment; however, achieving high accuracy while maintaining computational efficiency and interpretability remains a significant challenge. Even though knowledge distillation can improve overall efficiency in most cases, relying entirely on the teacher(s) can lead to poor predictions. This thesis presents a novel framework that integrates explainable artificial intelligence with an efficient knowledge distillation pipeline for wound segmentation. A dual-teacher knowledge distillation strategy is employed to transfer complementary representations from heterogeneous high-capacity models to a lightweight student network. To enhance the reliability of knowledge transfer, a confidence-aware knowledge filtering mechanism is introduced, which selectively guides the student model using only high-confidence predictions from the teachers. In addition, a lightweight yet effective student architecture, termed SE-HybridConv- Tiny-UNet, is proposed. This model incorporates hybrid convolutional operations and channel attention mechanisms to improve feature representation while maintaining a low parameter count. Extensive experiments conducted on wound segmentation datasets demonstrate that the proposed framework achieves competitive performance compared to larger models, while significantly reducing computational complexity. Furthermore, explainability is incorporated using GradCAM++, enabling visual validation of model decisions and ensuring that predictions are based on clinically relevant regions. The consistency observed between validation and test-time explanations highlights the generalization capability and reliability of the proposed approach. Overall, this work establishes an effective balance between accuracy, efficiency, and interpretability, making it suitable for deployment in real-world, resource-constrained healthcare environments.
Open Access
Probabilistic security mapping of large language model integrations via stochastic Petri Nets
(BRAC University, 2026) Mohammad, Zaber; Ferdous, Md Sadek; Sadeque, Farig Yousuf; Department of Computer Science and Engineering
Large Language Models (LLMs) are becoming increasingly popular for use in modern software systems. However, with increasing popularity, newly introduced security risks have emerged while integrating LLMs in a software system. These security gaps do not align with the traditional cybersecurity framework. To address it, this study specifically focuses on modeling three distinct related threats: prompt injection, context extraction, and Denial of Service (DoS) by resource exhaustion. First, the research maps these three LLM security aspects with the traditional CIA triad (Confidentiality, Integrity, Availability) and maps the system assets with corresponding justifications to show exactly what component of a system is at risk during these specific attacks. After that, the research investigates three distinct and independent threat models across the LLM architecture. First, Prompt Injection is analyzed at the input processing layer to mathematically evaluate Defensive Depth theory. Second, Data Exfiltration is evaluated during output scanning to formalize the Temporal Defense theory. Finally, a Denial of Service (DoS) attack is modeled to validate the Saturation theory. To transition from theoretical risk to measurable impact, an independent threat model is developed using Petri Net diagram to simulate these distinct stages of the LLM pipeline. Mathematical analysis is then conducted using a Continuous-Time Markov Chain (CTMC) and finite queuing theories. Specifically for the DoS evaluation, the adversarial arrival rate (λ) and system processing bottleneck (ρ) are modeled to measure the queue wait times and resource depletion. Across all three threat vectors, Monte Carlo validation is used to ensure the theoretical mathematical calculations match the simulated reality. The result provides a formalized mathematical baseline for each independent vulnerability. The findings demonstrate the exact architectural trade-offs to implement input-layer defensive depth, the temporal cost for output sanitization, and the critical threshold where system queues saturate and drop legitimate requests during a DoS attack. These insights help developer to design more resilient, optimized, and mathematically verifiable security architecture for deployed LLM applications.
Open Access
Unlearning to protect: a distilled reinforcement learning framework with privacy-preserving feature unlearning and XAI for IoT security
(BRAC University, 2025-11) Hasan, MD. Nahid; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Botnets pose a significant cybersecurity threat, enabling attacks such as DDoS, data theft, and service disruptions on IoT devices. These devices often lack built-in botnet traffic filtering, leaving them highly exposed. Existing AI-based solutions improve detection capabilities but have limitations: (i) they are too heavy for IoT deployment, and (ii) they lack unlearning capabilities to forget sensitive or outdated features without retraining. To address these challenges, we propose DiRLU, a lightweight, reinforcement learning driven framework, while ensuring privacy by selectively unlearning sensitive or outdated features without requiring retraining. The framework leverages knowledge distillation to transfer knowledge from a teacher model into a lightweight student model, with both models trained using A2C. A post-hoc unlearning mechanism modifies weights to remove targeted features, while restored features show negligible performance loss, confirming reversibility. Unlike many benchmark models that used only 5% of the BoT-IoT dataset, this research leverages 25%, allowing us to develop a strong teacher model. Both the teacher and student models were trained using the A2C reinforcement learning algorithm, achieving impressive results, with the student model achieving 99.60% accuracy and a 99.80% F1 score. To enhance transparency, we integrated Explainable AI (XAI), particularly LIME, which helps interpret the model’s decisions and identify the key features influencing its predictions. Additionally, DiRLU requires only 2,370 FLOPS, approximately 3.87× more efficient than the state-of-the-art model, highlighting its efficiency for edge deployment. DiRLU combines efficiency with privacy, aligning with GDPR standards (right to be forgotten) to provide practical IoT security solution. By combining knowledge distillation, feature unlearning and XAI, this research not only strengthens botnet detection but also sets new standards for security, interpretability, and data privacy in cybersecurity.
Open Access
Evidence-based workplace harassment guidance system: transforming #MeToo narratives into actionable knowledge through machine learning
(BRAC University, 2026-01) Kabir, Mashphey Bintey; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Workplace harassment remains a pervasive global issue, yet victims often lack ac- cess to evidence about what actually happens when people in similar situations take action. Traditional resources offer generic procedural advice without outcome data, leaving individuals to make consequential decisions with incomplete informa- tion. This study presents an evidence-based workplace harassment guidance system that transforms 15,835 #MeToo narratives into personalized, actionable guidance grounded in documented outcomes from comparable cases. The system addresses a fundamental information asymmetry: while organizations accumulate knowledge about harassment cases, individual victims rarely know what outcomes others in similar situations experienced. By analyzing patterns across thousands of documented experiences, the system identifies a user’s specific vulner- ability profile, retrieves semantically similar historical cases, and presents evidence- based guidance including proven successful action sequences, outcome statistics, and high-impact actions that correlate with positive results. The guidance generation pipeline employs SimCSE-BERT embeddings for semantic similarity, multi-label vul- nerability detection leveraging these embeddings to identify seven co-occurring risk factors, and outcome pattern analysis across fifteen outcome categories—supported by BERT sentiment analysis (98.1% accuracy) and HDBSCAN clustering for eval- uation. The empirical analysis reveals sobering realities: negative outcomes predominate across all vulnerability types, institutional inaction occurs in 16.9% of cases, and harasser accountability remains rare at 3.6%. Rather than offering false reassur- ance, the system presents these evidence-based statistics to enable informed decision- making. Professional evaluation with eleven practitioners from HR, legal, mental health, and other sectors—91% with direct harassment case experience—validates that the guidance meets practical standards for appropriateness (M=4.09/5), use- fulness (M=4.09/5), actionability (M=3.91/5), and safety (M=3.73/5), with 91% endorsement and 73% rating the approach superior to typical harassment resources. This research demonstrates that machine learning can provide meaningful support for sensitive domains when developed with rigorous professional validation and com- mitment to user safety.
Open Access
A lightweight time-series analysis model through multi-teacher knowledge distillation for food price forecasting
(BRAC University, 2026-01) Zaman, Shifat; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Accurate food price forecasting is critical for food security planning, particularly in developing nations like Bangladesh where price volatility can significantly im-pact vulnerable populations; however, existing deep learning approaches for time series forecasting often require substantial computational resources, limiting their deployment in resource-constrained environments. This study presents a novel multi-teacher knowledge distillation framework that transfers knowledge from an ensem- ble of three distinct teacher architectures—DLinear, PatchTST, and N-BEATS— trained on World Food Programme (WFP) Bangladesh commodity price data from the Dhaka Division to compact student models (MLP, GRU, KAN) through a multi- component distillation loss comprising prediction-level matching, feature-level alignment, and price-difference learning, with an uncertainty-weighted mechanism that focuses training on confident teacher predictions while dynamically weighting teacher contributions based on validation performance. Experimental evaluation on four food commodities (Lentils, Oil, Rice, and Wheat flour) with a 6-month input window demonstrates that the proposed approach achieves a Mean Absolute Error (MAE) of 1.959 BDT/unit with a Mean Absolute Percentage Error (MAPE) of only 3.73%, representing a 37% improvement over the supervised learning baseline, 69% improvement over traditional ARIMA, and 81% improvement over LSTM baselines, with the three-teacher ensemble distilled to an MLP student achieving the best results and outperforming all single-teacher and two-teacher configurations. The resulting student model requires only 200K parameters (compared to over 1M in the teacher ensemble) and achieves inference in sub-millisecond time on standard CPU hardware without GPU acceleration, enabling deployment in humanitarian field offices with limited computational infrastructure. This study contributes a reproducible, configuration-driven framework for knowledge distillation in time series forecasting, demonstrating that sophisticated ensemble-level accuracy can be achieved with lightweight models suitable for resource-constrained field deployment in food security applications.
Open Access
Integrated sensing and communication for next generation wireless systems
(BRAC University, 2026-01) Zabin, Rifat; Alam, Md. Golam Rabium; Department of Computer Science and Engineering
Wireless communication systems inherently interact with their surrounding physi- cal environment, causing transmitted signals to carry information beyond the data they are intended to convey. This observation forms the basis of Integrated Sensing and Communication (ISAC), where sensing and communication are jointly sup- ported using the same wireless infrastructure. Despite its promise, realizing ISAC in practical systems—particularly with commodity Wi-Fi—remains challenging due to high-dimensional channel representations, compute intensive channel sounding, hardware imperfection, and strong variability across environments. This thesis ad- dresses these challenges by developing practical, learning-driven ISAC frameworks that explicitly exploit the structural properties of wireless channels across time, fre- quency, and space. The thesis is organized around two complementary thrusts. First, communication-aided sensing is realized through PULSE , a lightweight framework that transforms raw Channel Frequency Response (CFR) measurements into compact, physics-aware temporal representations for sensing inference and keep- ing it generalized across domains. Second, sensing-aided communication is enabled through ELF , a scalable channel feedback framework that reformulates multi- antenna channel reporting as a structural inference problem, by feeding back only a small number of representative subchannel embeddings. Experiments show that PULSE achieves over 99% sensing accuracy across diverse activities while re- ducing the effective input dimensionality by approximately 85% and maintaining low inference latency suitable for real-time edge deployment. PULSE generalizes to unseen environments and devices using as little as 5 seconds worth of labeled data, outperforming state-of-the-art Wi-Fi sensing frameworks under domain shifts. On the other hand ELF reduces feedback overhead by up to 96% relative to dense sub- carrier reporting, while preserving near-identical communication reliability. Com- pared to standard IEEE 802.11ax explicit feedback, ELF achieves up to a 25× reduction in feedback size across wide bandwidths, and its feedback cost remains effectively independent of antenna count, enabling scalable operation in large Multiple-Input Multiple-Output (MIMO) systems. Extensive simulations and real- world Wi-Fi testbed evaluations demonstrate that the proposed approaches achieve strong sensing accuracy, robust cross-domain generalization, and substantial reduc- tions in communication overhead. Collectively, this thesis establishes a unified and deployable ISAC framework for commodity Wi-Fi systems, highlighting the role of channel structure in enabling efficient and reliable sensing and communication under real-world constraints.
Open Access
Real-time scene description and interpretation using zero-shot learning and prompt-engineered vision-language models
(BRAC University, 2024-08) Apurba, Md Shifatul Ahsan; Anwar, Md. Tawhid; Department of Computer Science and Engineering
Real-time scene description and interpretation are essential for diverse applications such as surveillance, interactive media, and automated video analysis. However, most existing methods rely heavily on large-scale labeled datasets, thereby limiting their adaptability in dynamic or previously unseen scenarios. In this work, we propose a novel mixed-model framework that integrates Vision-Language Models (VLMs), Large Language Models (LLMs), and lightweight object detection networks (e.g., MobileNet-SSD) through advanced prompt engineering. By leveraging zeroshot learning, our approach generates contextually rich scene descriptions without requiring domain-specific or task-specific retraining. The prompt engineering component reduces sensitivity to subtle linguistic variations, enhancing robustness across diverse input formulations. Furthermore, the lightweight detector ensures real-time performance, making the framework suitable for resource-constrained environments. To address ethical and fairness considerations, we incorporate bias mitigation strategies that limit the propagation of harmful stereotypes from large-scale pretraining data. Experimental evaluations on multiple open-domain scenarios demonstrate that our system offers reliable and efficient scene interpretation, maintaining high accuracy in challenging conditions where traditional supervised techniques often fail. This research paves the way for more flexible, scalable, and responsible visionlanguage systems capable of operating effectively in real-world, zero-shot contexts.
Open Access
WHISNER-BN: parameter-efficient end-to-end spoken named entity recognition for low-resource languages with morphology-aware alignment
(BRAC University, 2025-12) Imran, Shah; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Named entity recognition from speech remains underdeveloped for low-resource languages such as Bengali, despite its importance for voice search, conversational AI, and accessibility. This thesis investigates Bengali spoken NER through three paradigmsdiscriminative structured prediction, generative multi-task learning, and multimodal instruction-tuned approaches-with the primary contribution whisNer-bn, a parameterefficient architecture employing Low-Rank Adaptation of Whisper encoders (7.1% trainable parameters), BiLSTM contextual encoding, and Conditional Random Field decoding with explicit BIO constraints. To enable end-to-end training, we introduce bnSpAligner, a morphology-aware forced alignment algorithm achieving 78% tokenlevel accuracy for Common Voice and 83% for SUBAKKO through adaptive thresholding and phonetic equivalence classes, and BSSC-Annotator, a multi-agent framework leveraging cross-lingual transfer to produce 50.57 hours of annotated Bengali speech comprising 38,504 entities across 36,237 utterances at less than 10% of manual annotation cost. Evaluated on 5,000 samples from combined test partitions using models trained on 20% of available data (4,989 samples), whisNer-bn achieves 0.681 F1, outperforming generative multi-task approaches by 15.6 percentage points and cascaded ASR-NER pipelines by 5.8 percentage points. The results demonstrate that discriminative structured prediction with joint acoustic-linguistic modeling provides superior inductive biases for entity recognition in morphologically complex languages, establishing the first systematic benchmark and transferable methodology for Bengali spoken NER with implications for thousands of underserved languages.
Open Access
Causal inference in depression: understanding beyond correlation
(BRAC University, 2025-10) Shoumo, Syed Zamil Hasan; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Depression remains one of the most pressing mental health concerns worldwide, intensified further by the socioeconomic and psychological impacts of the COVID-19 pandemic. Understanding the underlying mechanisms that contribute to depressive symptoms has therefore become a major research priority. While traditional statistical and machine learning models have been effective in identifying associations between risk factors and depression, they often fail to distinguish correlation from causation. Explainable Artificial Intelligence (XAI) methods, such as SHAP and LIME, have improved transparency by revealing which features influence model predictions; however, they remain fundamentally correlational and do not provide insight into the true causal pathways that drive depressive outcomes. To address this limitation, this research integrates machine learning, explainable AI, and causal inference to explore the causal factors behind depression among the Bangladeshi population during the COVID-19 pandemic. Using XGBoost for predictive modeling, the study first evaluates the relative importance of features through gain-based measures and SHAP value interpretation. Subsequently, a causal inference framework is constructed following Judea Pearl’s principles to identify and estimate direct causal effects using the backdoor adjustment method with a generalized linear model estimator. Finally, a combined feature-selection pipeline is developed that retains causally significant variables and iteratively removes weakly correlated ones to test their joint predictive strength. The results reveal that while several factors exhibit high correlation and feature importance in black-box models, only a subset demonstrates genuine causal influence on depressive outcomes. This distinction underscores the importance of causal reasoning in mental health analytics. Overall, the study establishes that integrating causal inference within predictive frameworks not only enhances interpretability and trustworthiness but also provides a clearer understanding of which factors can truly influence and potentially mitigate depression.
Open Access
Chained semantic retrieval for rare disease and gene identification using clinical phenotype
(BRAC University, 2025-10) Saiful, Md.; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Accurate identification of rare diseases and associated genes from patient phenotypic data represents a critical challenge in precision medicine and genomic research. Current diagnostic approaches face substantial limitations in scalability, interpretability, and real-time processing when integrating heterogeneous phenotypic and genotypic databases. This study presents a novel artificial intelligence-driven system employing chained semantic retrieval and knowledge graph integration to identify rare diseases, genes, and Human Phenotype Ontology (HPO) terms from patient phenotype descriptions. The methodology leverages sentence transformers (based on Bidirectional and Auto-Regressive Transformers) for generating vector embeddings, Facebook AI Similarity Search (FAISS) for efficient similarity computation, and a fine-tuned Llama 3.2 model integrated with a Retrieval-Augmented Generation (RAG) pipeline. The system implements a tiered scoring mechanism that chains retrieval across Human Phenotype Ontology, disease, and gene databases to progressively refine predictions through contextual enhancement. Evaluation on 50 patient phenotypes with confirmed Duchenne Muscular Dystrophy diagnosis, consisting of 30 true positive and 20 true negative cases, demonstrated strong performance with 28 true positives, 17 true negatives, 2 false negatives, and 3 false positives in top-ten retrieval results. The system achieved 90 % overall accuracy, 93.3 % recall, 85 % specificity, 90.3 % precision, and an F1-score of 91.8 %, with an average computational efficiency of 3.2 seconds per response. The proposed framework effectively addresses critical gaps in cross-database integration while maintaining interpretability through tiered confidence scoring for clinical decision support applications.
Open Access
OmniNet: a hybrid deep learning framework for robust semantic segmentation
(BRAC University, 2025-10) Ifty, Fahim Ahmed; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Semantic segmentation empowers computers to interpret visual scenes in a structured and meaningful manner, accurately delineating the precise boundaries of every object in an image at the pixel level. However, existing CNN-based models struggle with long-range context, transformer-based approaches are computationally heavy and often miss local detail, and many hybrid designs neglect explicit multi-scale features or refined attention, while also suffering from class imbalance and noisy annotations. This research proposes an encoder–decoder architecture that combines a ResNet152 backbone, Atrous Spatial Pyramid Pooling (ASPP), and a Transformer Refine Block, leveraging the strengths of convolutional features and lightweight self-attention to capture both multi-scale semantics and long-range dependencies. In the decoder, a UNet-style upsampling path with skip connections and CBAM attention preserves spatial detail and refines boundaries, yielding coherent pixel-level predictions across scales. The training pipeline applies label cleaning and robust augmentation (MixUp, CutMix) and optimizes a composite loss (Lovasz–Softmax, Dice Loss, Boundary Loss) to counter label noise and the long-tail distribution. Evaluated on the CamVid benchmark, our method achieves 86.53% mean IoU and 95.99% pixel accuracy, outperforming recent efficient residual attention networks while prioritizing segmentation quality over inference speed. These results demonstrate that combining multi-scale convolutional features with lightweight self-attention and boundary-aware optimization delivers state-of-the-art accuracy under noisy labels and severe class imbalance in urban scene segmentation.
Open Access
A data-driven institutional study of academic performance: exploring course, structure, and environment in higher education
(BRAC University, 2025-10) Shakil, Arif; Kazi, Sadia Hamid; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Student performance reflects the intersection of academic, structural, and environmental factors rather than individual effort alone. This study conducts a largescale, data-driven analysis of institutional records from BRAC University to examine how these factors shape outcomes across time, delivery modes, and disciplines. Nine hypotheses were tested, encompassing the effects of online, hybrid, and inperson learning environments; the COVID-19 pandemic; course-level contributions to CGPA; prerequisite–core alignment; class size; Residential Semester (RS) contexts; high-failure course patterns; and inter-departmental performance differences. Using Pearson correlation, regression, ANOVA, Kruskal–Wallis, Welch’s t-tests, and mixed-effects models, the study identifies clear structural trends: RS participation strongly correlates with higher and more consistent GPA; class size shows weak, context-dependent effects; and persistent high-failure rates cluster in STEM gateway courses. Departments differ systematically—humanities and law programs maintain higher GPAs, while technical disciplines show greater variance due to assessment rigor and prerequisite dependency. These findings reveal that academic outcomes are institutionally patterned, not random. They underscore the need for data-informed curriculum design, departmental benchmarking, and early-risk intervention frameworks to promote equity, quality, and resilience in higher education.
Open Access
Designing a Bangla conversational AI agent for maternal health using model context protocol
(BRAC University, 2025-09) Muntasir, Fahim; Noor, Jannatun; Department of Computer Science and Engineering
Maternal health in Bangladesh faces persistent challenges, including limited access to skilled providers, informational gaps, and stigma around perinatal mental health. Prior studies show that digital health tools remain constrained by generic content, English-dominated design and neglect of maternal mental health, limiting trust and engagement. We present Baby and Me; the first Model Context Protocol-enabled agentic AI system designed for maternal health in Bangladeshi contexts. The system delivers personalized, empathetic guidance in Bangla and Banglish by combining retrieval-augmented generation with a clinically curated knowledge base, web search, and conversational memory. A survey with 72 women revealed frequent worries about miscarriage, anxiety, and mood changes, highlighting the need for empathetic, accessible support. Evaluation of our prototype showed high contextual accuracy, low hallucination, and strong user satisfaction, with participants valuing empathy and trust while requesting greater personalization. Our findings extend human-AI interaction research by demonstrating how culturally grounded, agentic AI can serve not only as an informational tool but also as a relational companion, offering design insights for equitable health technologies. It paves the way for scalable interventions in low-resource settings, with future directions to enhance maternal outcomes. The chatbot can be accessed in this link.
Open Access
Industry perspectives on Kubernetes: a multi-method empirical analysis of adoption drivers, usage patterns, and ecosystem activity
(BRAC University, 2025-09) Taposhi, Umme Jannat; Haque, S M Taiabul; Department of Computer Science and Engineering
Kubernetes has rapidly become the backbone of cloud native computing, yet we still know relatively little about how it is adopted and practiced in day-to-day software engineering work. Despite its central role in modern infrastructure, systematic empirical research that captures both practitioner experiences and open-source evidence has been largely absent. To address this gap, we conducted a mixed-method study combining 22 semi-structured practitioner interviews, survey responses from 31 professionals, and an analysis of 1,429 public repositories. Together, these sources provide a comprehensive view of Kubernetes adoption, usage, and challenges. The interviews informed the design of the survey, which in turn guided our analysis of repositories. By linking practitioner perspectives with the technical evidence from repositories, we were able to observe which practices appeared frequently, which were less common, and which showed variation between the two sources. Our findings highlight the widely shared practices as well as gaps where aspirational best practices are not consistently reflected in real-world repositories. We also highlight ongoing five key challenges that illustrate how Kubernetes remains a maturing ecosystem rather than a fully developed technology. Taken together, this work contributes to a deeper understanding of Kubernetes in practice. For practitioners, it offers lessons for navigating adoption and maintenance, while for researchers, it provides a unique empirical dataset that links qualitative insights with quantitative repository evidence.
Open Access
A deep learning framework for arsenic skin disease detection leveraging dual-teacher knowledge distillation with a depthwise-separable convolution and KAN-based lightweight student model
(BRAC University, 2025-12) Mehedi, Md Humaion Kabir; Mridha, Muhammad Firoz; Department of Computer Science and Engineering
Arsenic poisoning in groundwater poses a major public health issue in Bangladesh. Chronic arsenic poisoning often leads to arsenicosis, a chronic disease characterized by cutaneous effects, including melanosis, leukocomelanosis and keratosis. Timely treatment of these lesions on the skin is important since prompt diagnosis and timely action are taken within the medical profession. However, this is challenging in rural areas. In most areas, there are no trained dermatologists, diagnoses tend to be subjective and there may be inadequate healthcare facilities. The proposed study involves the development of an explainable and lightweight deep learning framework for the automatic identification of skin diseases caused by arsenic, which will be computationally efficient and interpretable on a clinical scale. This system uses the dual-teacher knowledge distillation (KD) approach, in which two high-capacity models, InceptionV3 and Xception+InceptionM, are used to distill discriminative and contextual information to a small student model called Inception-Residual- KANNet (IR-KANNet). ArsenicSkinImageBD dataset was trained and validated using a stratified data split and 5-fold cross-validation to ensure class balance and model stability. The final evaluation found accuracy 0.9692, precision 0.9610, recall 0.9867 and F1-score 0.9737 on both infected and not-infected cases using stratified data split and a factor of 78.39% reduction in parameters over the teacher 1 and 35.29% in teacher 2 networks. These results indicate the appropriateness of the model for use in low-resource healthcare settings by providing convenient and reliable AI-based screening for arsenicosis detection. Grad-CAM++ and LIME make the model explainable and the resulting transparent heatmaps are consistent with the clinical regions in terms of lesions. This study is relevant for developing interpretable, efficient and domain-flexible medical AI models. It also provides a basis for further study of explainable knowledge distillation and edge-deployable solutions for the diagnosis of other dermatological disease
Open Access
GAFFDNet: GTE-based adaptive fusion and full-scale decoding for referring image segmentation
(BRAC University, 2025-11) Chowdhury, Bushra Rafia; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
Referring Image Segmentation (RIS) requires a precise understanding of both complex visual scenes and natural language expressions to segment target objects accurately. Despite recent advancements, traditional transformer-based methods often face challenges with ambiguous expressions, weak vision-language alignment, and limited contextual reasoning. These limitations hinder the ability to capture nuanced interactions between language and visual context, particularly in cluttered or complex scenes. This research proposed GAFFDNet, an innovative paradigm for referring image segmentation that addresses these challenges. A multi-stage contrastively trained model (GTE) is employed to produce semantically rich and discriminative textual embeddings. A dynamic multimodal fusion module is designed to adaptively integrate visual and linguistic information, which allows the network to focus on the most relevant cues. A full-scale or multi-scale feature aggregation based decoder is incorporated to facilitate dense information exchange across different scales, thereby enhancing both fine-grained spatial detail and global context understanding. Extensive evaluations conducted on three standard benchmark datasets, RefCOCO, RefCOCO+, and G-Ref demonstrate that GAFFDNet consistently outperforms existing competitive methods. These results validate the effectiveness of the proposed approach in achieving precise, context-aware segmentation aligned with complex referring expressions.

Browse

Recent Submissions