Bishop’s University

Dr. Mebarka Allaoui Post-Doc at Bishop’s University: Edge-based AI Methane Gas Leak Detection, Prediction, and Monitoring for a Cleaner Environment

Dr. Mebarka Allaoui is a postdoctoral researcher in computer science at Bishop’s University, where she conducts foundational research in machine learning with broad applications across diverse domains. She chose Bishop’s University for her postdoctoral training because of its supportive research environment and the strong alignment between her research interests and those of Dr. Rachid Hedjam. The institution offers an ideal balance between foundational machine learning research and applied collaborations, allowing her to develop rigorous theoretical methods while testing them in diverse real-world scenarios. Her work focuses on developing principled and scalable machine learning methods capable of extracting meaningful structure from complex, high-dimensional data while remaining robust and efficient in real-world settings.

She began her academic journey at the University Kasdi Merbah in Algeria, where she earned a bachelor’s degree in Information Systems in 2016, followed by a master’s degree in Industrial Computer Science in 2018. She completed her Ph.D. in Computer Science at the same institution in 2024. Her doctoral research investigated advanced embedding techniques and representation learning methods for high-dimensional data. Throughout her studies, she developed a strong interest in the theoretical and algorithmic foundations of machine learning, leading to expertise in dimensionality reduction, embedding learning, clustering, coreset selection, and graph-based modeling, including graph neural networks (GNNs).

Her path toward research was shaped early in life. Inspired by her mother’s encouragement to cultivate curiosity and a love of learning, she aspired from a young age to become a scientist and contribute meaningful ideas to society. During high school, she discovered her passion for computer science, which naturally guided her toward university studies in the field. Along the way, she benefited from the mentorship of dedicated professors who nurtured her intellectual growth. Dr. Messaoud Mezati significantly strengthened her appreciation for academic research, while her doctoral supervisor, Dr. Mohamed Lamine Kherfi, played a central role in shaping her research development. More recently, her postdoctoral supervisor at Bishop’s University, Dr. Rachid Hedjam, has provided continuous support and an intellectually stimulating environment that has enabled her to further develop as an independent researcher.

Her research is driven by fundamental challenges in machine learning, particularly the problem of learning compact, informative, and structure-aware representations from complex data. Issues such as dimensionality reduction, embedding learning, coreset selection, and clustering are central to modern machine learning systems and arise across numerous application domains. She is especially interested in graph-based and manifold-aware approaches, which capture relationships and intrinsic geometry often overlooked by traditional models.

Dr. Allaoui’s work contributes to advancing knowledge in the field by proposing principled frameworks that emphasize structure preservation, scalability, and robustness. By integrating concepts from manifold learning, graph neural networks, and data summarization techniques such as coreset selection, her research seeks to improve how models learn from large, heterogeneous, and high-dimensional datasets. In the long term, she hopes her work will contribute to the development of more reliable, efficient, and adaptable machine learning systems.

Her current research projects continue to explore foundational machine learning problems, including representation learning, graph-based modeling, and data-efficient learning strategies. While her methods are applied in varied contexts—such as financial transaction analysis in a previous project funded by Mastercard and Mitacs, and edge-based sensing systems in a current TinyML project on gas leak detection funded by Pergamon Perceptive Technologies in coordination with Mitacs—her primary contribution remains methodological. Advances in dimensionality reduction, embedding learning, coreset selection, and graph-based clustering are essential for ensuring that such applications are scalable, robust, and deployable, particularly in settings characterized by high dimensionality, limited data, or constrained computational resources.

Looking ahead, Dr. Allaoui envisions a future in which machine learning systems increasingly prioritize principled representations, efficiency, and interpretability. As datasets continue to grow in size and complexity, approaches such as graph neural networks, coreset selection, and manifold-aware learning will play an increasingly significant role. She believes the field will continue evolving toward models that are not only accurate but also scalable, adaptable, and grounded in strong theoretical foundations, enabling responsible and effective deployment across diverse domains.

—

Joannie St-Germain B.Sc. ’16, M.Sc. (she/her/elle)
RESEARCH OFFICER
OFFICE OF RESEARCH AND GRADUATE STUDIES
819-822-9600 x 2242

Similar Posts