Please join us for the following teaching presentation by a candidate for the faculty position in the Department of Computer Science.
Presenter: Dr. Aaron Williams
Teaching Talk: Balanced Binary Search Trees
For the Zoom link, please contact grad.compsci@lakeheadu.ca.
FOBA Speaker Series presents Dr. Aydin Farrokhi, Assistant Professor, Business Analytics and Information Systems at Lakehead University.
Friday, May 1
12:10 to 1 p.m.
ATAC 1007 or Zoom (https://lakeheadu.zoom.us/j/97324803134)
Crises rarely announce themselves—they hide in organizational data. This study shows how AI agents detect crises early, turning communications into signals of disruption. Drawing on situational crisis communication theory, we demonstrate how machine learning enables firms to sense and act—transforming management from reaction to foresight.
Please join the Department of Computer Science for the upcoming thesis defense:
Presenter: Protiva Arafin
Thesis title: Evidence-Grounded Clinical Pharmacogenomics Question Answering System Using Large Language Models and Hybrid Retrieval Augmentation
Abstract: Pharmacogenomics (PGx) is very important for personalized medicine since it helps doctors choose the right drugs and doses based on a person’s genetic makeup. But the growing amount and complexity of PGx data, as well as the requirement to understand clinical recommendations, make it harder to make good decisions. This study puts forward a data-driven clinical decision support framework that combines large language models (LLMs) with hybrid retrieval-augmented generation (RAG) to enhance the responding of pharmacogenomic questions.
The framework assesses two contemporary LLMs, Meta-LLaMA-3.1-8B-Instruct and Qwen3-8B, through various configurations, encompassing base models, Low-Rank Adaptation (LoRA) fine-tuning, and hybrid RAG-based methodologies. The structured pharmacogenomics data from CPIC and the clinical guideline information from ClinPGx are combined to make a huge dataset. To make it easier to find and use in models, the data goes through procedures including merging, cleaning, normalizing, and converting to JSONL format.A hybrid retrieval approach is aimed to enhance factual grounding by integrating lexical filtering with semantic similarity through sentence embeddings. This research use both automatic metrics and manual checks to rate the models on their correctness, relevance, completeness, and clarity. The results reveal that Qwen works well as a basic model, and that LLaMA gets much better when it is used with RAG and LoRA, giving answers that are more aware of the context and therapeutically useful. Fine-tuning alone doesn’t always work, which shows how limited it is to only use parametric data. The results show that accuracy in clinical settings needs to be backed up by consistency, relevance, and evidence.
This study demonstrates that employing retrieval methods alongside parameter-efficient fine-tuning enhances the reliability and utility of LLM-based systems in clinical environments. The proposed methodology establishes a scalable framework for the development of trustworthy AI-driven solutions in pharmacogenomics and healthcare decision support.
Committee Members:
Dr. Abedalrhman Alkhateeb (supervisor, committee chair), Dr.
Md Moniruzzaman (co-supervisor), Dr. Saad B. Ahmed, Dr. Malek Alsmadi (Electrical & Computer Engineering)
Please contact grad.compsci@lakeheadu.ca for the Zoom link. Everyone is welcome.
Please join the Department of Computer Science for the upcoming thesis defense:
Presenter: Alireza Rajoli Nowdeh
Thesis title: Domain Adaptation for Robust WiFi Sensing: Empirical Analysis of Domain Shift and Class Imbalance in WiFi CSI-based Human Activity Recognition
Abstract: WiFi-based human activity recognition (HAR) has emerged as a promising device-free sensing technology for smart homes, healthcare monitoring, and ambient assisted living. However, these models suffer from performance degradation when applied to new environments because of domain shift that caused by changes in room layout, multipath propagation, and line-of-sight (LOS) versus non-line-of-sight (NLOS) conditions. This thesis addresses the robustness gap by extending Dual Adversarial Network for Human Activity Recognition (DA-HAR) with two components: a Conditional Domain Adversarial Network (CDAN) for aligning joint feature and prediction distributions and a class-weighted learning strategy to mitigate the effects of class imbalance.
The proposed framework is evaluated by a public WiFi CSI dataset with 12 fine-grained activities which are collected in three environments (corridor, office, classroom) under LOS and NLOS conditions. Comprehensive experiments are conducted using three transfer scenarios (E1,E2→E3 / E1,E3→E2 / E2,E3→E1). The results indicate that the enhanced DA-HAR consistently outperforms both the original DA-HAR and the source-only baseline achieving up to 7.1% absolute and 17.2% relative accuracy improvements. The confusion matrix shows that DAHAR + CDAN reduces misclassification among fall-related and locomotion activities, meaning that it preserves class structure under domain shift better.
To isolate the effect of class imbalance, an experiment is performed on 20 random six-activity subsets with and without class-weighted loss. Class weighting improves accuracy from 0.6806 to 0.7112 and increases CDAN improvement from 5.3% to 8.7%, which means that class imbalance negatively impacts both the source model and the adaptation process. However, fall activities depend on the scenario, and bending is still difficult, showing remaining challenges.
This work provides empirical evidence that robust domain adaptation for WiFi sensing both conditional alignment of class-specific distributions and explicit treatment of class imbalance. The proposed method offers a principled step toward reliable activity recognition in real-world environments.
Committee Members:
Dr. Ali Nazari Shirehjini (supervisor, committee chair), Dr. Abedalrhman Alkhateeb, Dr. Ehsan Atoofian (Electrical & Computer Engineering)
Please contact grad.compsci@lakeheadu.ca for the Zoom link. Everyone is welcome.
Please join the Department of Computer Science for the upcoming thesis defense:
Presenter: Elena Krikun
Thesis title: Causal Discovery and Treatment Effect Modeling in Breast Cancer
Abstract: Modeling breast cancer outcomes remains challenging because of extreme molecular heterogeneity and the inability of associative models, including those developed through traditional machine learning, to support counterfactual, intervention-based clinical reasoning. Building on recent advances in causal feature selection, multiomics variable selection, and individual treatment effect estimation, this thesis proposes a hybrid pipeline within a unified computational multiomics framework that integrates high-dimensional data with causal modeling to produce interpretable precision oncology models that extend beyond risk prediction.
The proposed pipeline was developed using the TCGA-BRCA cohort as the discovery set and validated on the independent retrospective METABRIC cohort to assess transportability. To address the curse of dimensionality, the framework applies Markov Blanket-based local causal discovery across seven data modalities and reduces more than 600,000 initial features to a sparse and stable causal core. This causal representation is then used for survival modeling (C-index = 0.8085, 5-year AUC = 0.8676) and individual treatment effect (ITE) estimation for chemotherapy, hormone therapy, and targeted therapy. External validation on METABRIC achieved a C-index of 0.7200 and a 5-year AUC of 0.7639, indicating moderate but clear transportability across cohorts and assay platforms. The final causal core confirmed the integration of clinical, proteomic, and epigenetic signals, and identified a long non-coding RNA as a structurally relevant driver.
The treatment-effect stage used treatment-specific arm definitions reconstructed from clinical records together with a robustness-oriented validation protocol. Chemotherapy showed the strongest and most stable beneficial treatment effect, most notably in the TNBC subgroup, where treatment-effect estimates remained consistently protective across estimators and overlap-adjusted variants. Hormone-therapy estimates showed a consistently protective direction in receptor-positive subgroup analyses, although the magnitude of the effect was attenuated under stricter overlap control, indicating residual confounding and limited positivity in the observational setting. Targeted therapy also showed a protective direction under most evaluated techniques, but given the very small number of treated patients and partial estimator disagreement, these effect estimates should be interpreted as exploratory.
Committee Members:
Dr. Abedalrhman Alkhateeb (supervisor, committee chair), Dr. Saad B. Ahmed, Dr. Maysa Yaseen (Electrical & Computer Engineering)
Please contact grad.compsci@lakeheadu.ca for the Zoom link. Everyone is welcome.
Please join the Department of Computer Science for the upcoming thesis defense:
Presenter: Yutao Zhou
Thesis title: An Integrated Framework for Art Image Novelty Detection and Ownership Tracing Using Deep Siamese Networks and Blockchain
Abstract: Digital artworks are increasingly distributed and shared across online platforms, raising critical challenges in both ownership verification and similarity detection. On one hand, it is difficult to establish secure, tamper-resistant records of artwork ownership in decentralized environments. On the other hand, identifying whether a newly submitted artwork is visually similar to an existing one remains a non-trivial task, especially under various artistic transformations such as style transfer, inpainting, and compositional editing. Vision Models can effectively compare image content, but they lack mechanisms to securely protect ownership. In contrast, blockchain technologies offer immutability, traceability, and decentralized data storage, yet lack the capability to evaluate visual similarity. These limitations highlight the need for an integrated solution that jointly addresses both visual similarity detection and secure ownership verification.
To resolve those issues, this thesis proposes a blockchain-based artwork verification system that integrates deep visual similarity detection with decentralized ownership registration. In the proposed framework, blockchain is used to register artwork ownership and store compact image feature payloads as immutable on-chain records, while offchain deep learning models extract visual embeddings and perform similarity matching. Multiple visual models are trained and evaluated under different distance metrics, loss functions, and Siamese architectures. To improve the practicality of on-chain storage, the extracted embeddings are projected into lower dimensions, quantized into compact payloads, and then analyzed for storage feasibility and matching performance. The blockchain component is further evaluated through experiments on payload storage, update cost, retrieval efficiency, and large-scale matching simulation.
Experimental results show that the ResNet50 model trained with NT-Xent loss and Euclidean distance achieves the best overall performance among the tested settings, while DeiT-small performs competitively at higher embedding dimensions. The results further indicate that quantized and compressed embeddings can significantly reduce blockchain storage cost while preserving most retrieval capability, although lower-dimensional embeddings increase the false-positive rate in the final similarity simulation.
Committee Members:
Dr. M. Mazhar Rathore (supervisor, committee chair), Dr. Moira MacNeil, Dr. Yong Deng (Software Engineering)
Please contact grad.compsci@lakeheadu.ca for the Zoom link. Everyone is welcome.
The Social Justice Studies Research Day Presentations will take place on April 22, 2026, at AT5041 (Thunder Bay Campus) and OA2020 (Orillia Campus), with a Zoom option available.
To receive the Zoom link, please email Admin.socialjustice@lakeheadu.ca.
This event features research and course-based presentations by graduate students on a range of social justice topics.
This session, titled "Capturing Intersectional Lived Experiences: Soundsourcing as Research Method" will be presented by Dr. Kathy Sanderson, an Associate Professor in Lakehead University's Faculty of Business Administration and Associate Director of the EPID@Work Research Institute.
Please register if you are interested in attending EPID Talks on Wednesday April 30, 2026 from 12:00 - 1:00 p.m. EST; participants can join in-person at Lakehead University in the CASES Building room FB 2023, or online through Zoom.
Please join the Computer Science department for the upcoming thesis defense:
Presenter: Jay Vinit Lunia
Thesis title: Trustworthy Learning with Graph-Aware Quantum-Classical Transformers for Hyperspectral Imaging and NLP Tasks
Abstract: This thesis develops a unified framework for trustworthy and efficient learning under constrained representations, with experiments spanning hyperspectral image understanding and natural language classification. The central premise is that representation constraints should be designed as explicit modeling choices rather than treated as unavoidable limitations. Following this premise, the thesis builds architectures that combine structured inductive bias, compact latent interfaces, and reliability-centered evaluation. For hyperspectral learning, the proposed methodology integrates patch-level graph construction with transformer attention to preserve local spatial coherence while still modeling long-range spectral interactions. This design is extended to multitask prediction by sharing a constrained backbone across classification and regression objectives. For compact hybrid modeling, the thesis introduces a low-dimensional quantum–classical token interface in which a variational quantum encoder acts as a structured bottleneck before lightweight transformer blocks. For reliability analysis, the thesis evaluates not only aggregate predictive accuracy, but also adversarial sensitivity, explanation stability, and cross-setting consistency under matched interface constraints. The empirical study covers multiple benchmark settings with heterogeneous spectral characteristics and class imbalance profiles, together with text classification tasks under fixed encoder conditions. Results show that the proposed constrained architectures achieve strong predictive performance while maintaining stable behavior across datasets and tasks. In hyperspectral experiments, the graph-aware transformer pipeline produces consistently high class-level and global metrics. In compact hybrid experiments, the quantum-classical interface remains competitive at low parameter budgets and reveals distinct robustness and attribution patterns when compared with a matched classical head. Overall, the thesis establishes a single integrated claim: principled representation constraints can improve reliability, interpretability, and deployment practicality without requiring a trade-off against competitive predictive quality. The work contributes an end-to-end blueprint for designing constrained yet trustworthy learning systems, including architecture design rules, evaluation protocols, and practical guidance for future extensions to stricter generalization tests and hardware-grounded hybrid inference.
Committee members:
Dr. Saad B. Ahmed (supervisor, committee chair), Dr. Garima Bajwa, Dr. Thangarajah Akilan (Software Engineering)
Please join the Computer Science department for the upcoming thesis defense:
Presenter: Mohammed Salman Khan
Thesis title: Crop Disease Analysis through Hyperspectral Images Using Deep Learning Models
Abstract: Modern precision agriculture increasingly relies on high-resolution Unmanned Aerial Vehicle (UAV) hyperspectral imagery to map diverse vegetation species and monitor complex crop health. However, processing these massively high-dimensional data cubes historically requires classical deep learning models with unsustainable computational bloat, such as heavy vision transformers or extremely deep convolutional networks. Furthermore, standard optimization pipelines routinely collapse when confronted with the complicated structural complexities of real-world agricultural datasets, which naturally feature severe class imbalances and highly overlapping spatial boundaries. This thesis directly attacks these critical computational and mathematical vulnerabilities by engineering ultra-lightweight, parameter-efficient hybrid quantum-classical architectures. By entirely replacing massive classical dense layers with a parameterized 4-qubit variational quantum circuit, this research demonstrates that quantum mechanics can natively and efficiently synthesize the highly complex, non-linear global dependencies required for accurate field classification.
To overcome the distinct spatial and spectral challenges of agricultural data, this work introduces two novel evolutionary frameworks. The first, the Quantum Patch-Graph Transformer (QPGF), mathematically preserves orthogonal crop row geometry by structuring spatial patches into row-normalized 4-nearest neighbor graphs, seamlessly fusing local graph attention with quantum global feature extraction. The second methodology is the Quantum Enhanced CNN-BiSpectralMamba-Quantum architecture, which actively bypasses standard memory bottlenecks by utilizing bidirectional Mamba state-space models to aggressively process continuous spectral sequences at linear complexity. Both architectures are stabilized by a custom Hybrid Cross-Entropy and Log-Cosh Dice loss function. This highly specialized optimization pipeline strictly forces the networks to penalize dominant staple crops and accurately map the topological boundaries of rare, minority vegetation.
Rigorous empirical validation on the highly imbalanced, 200-band, 30-class UAV-HSI-Crop dataset proves the absolute efficacy of these hybrid designs. The classical-quantum fusion drastically reduced the total trainable parameter count compared to state-of-the-art classical benchmarks. Despite this incredibly lightweight computational footprint, the QPGF established a robust baseline of 81.92% Overall Accuracy, while the advanced Quantum Enhanced CNN-BiSpectralMamba achieved a highly competitive peak of 84.83% Overall Accuracy and 82.07 Kappa score. Ultimately, this thesis proves that fusing targeted classical spatial-sequence extractors with quantum state entanglement provides a mathematically elegant, highly scalable, and resource-efficient diagnostic engine for the future of precision agriculture.
Committee members:
Dr. Saad B. Ahmed (supervisor, committee chair), Dr. Abedalrhman Alkhateeb, Dr. Ehsan Atoofian (Electrical & Computer Engineering)
