CSE Seminars - School of Computational Science & Engineering - Faculty of Science

News

We hope to see you all in our upcoming events. Please see seminar announcements for locations.

-Seminar chairs

CSE Seminars, The Goals

The Scientific Computing Seminar is an interdisciplinary, student-oriented event which serves as a venue for dissemination of information about various aspects of scientific computing such as:

applications of scientific computing to various disciplines, including some non-traditional ones,
technical aspects of high-performance computing,
tutorials concerning practical aspects of computing (hardware/software issues, code development, parallelization, debugging, etc.).

The selection of topics highlights both the breadth and depth of research in scientific computing at McMaster.

All Fall 2025 Seminars will be held from 12:30pm – 1:30pm in Hamilton Hall 410 on the following dates.

Past Seminars

Mar 18, 2026 - Numerical Approaches for Variational Data Assimilation Using Multigrid Methods, Adaptive Wavelets, and Fourier Neural Operators

Speaker: Hamidreza Moazzami
When: Mar 18, 2026
Time: 12:30 -13:20pm
Where: Hamilton Hall, 410

Title: Numerical Approaches for Variational Data Assimilation Using Multigrid Methods, Adaptive Wavelets, and Fourier Neural Operators

Bio: Hamidreza Moazzami is a PhD candidate in the School of Computational Science and Engineering at McMaster University. His research interests include PDE-constrained optimization problems, numerical analysis, and data assimilation. His previous work has involved time series analysis, Kalman filtering, and stochastic differential equations. His current research focuses on variational data assimilation and the development of methods to accelerate it using adaptive wavelets, multiscale and spectral analysis, and machine learning.

Abstract:

Variational data assimilation plays a central role in many scientific and engineering applications, particularly in areas such as weather prediction and geophysical modelling. Data assimilation combines sparse observational data with mathematical models governed by partial differential equations (PDEs) to produce improved estimates of the system state. However, solving the resulting optimization problems can be computationally expensive due to the high dimensionality of the underlying systems.

In this talk, we present approaches to accelerate Hessian-based variational data assimilation by leveraging the multigrid method, the adaptive wavelet collocation method, and the Fourier neural operator (FNO). By exploiting the inherent multiscale structure of PDE-constrained optimization problems, these methods enable more efficient computations and improved scalability.

The talk will discuss the mathematical formulation of the problem, the proposed acceleration strategies, and preliminary results demonstrating their effectiveness in data assimilation applications.

Mar 4, 2026 - A Transformer-based Reconstruction Framework for Sparse Hierarchical Clustering

Speaker: Dr. Pratheepa Jeganathan
When: Mar 4, 2026
Time: 12:30 -13:20pm
Where: Hamilton Hall, 410

Abstract: Sparse hierarchical clustering (SHC) traditionally identifies cluster partitions and a global weight vector by maximizing a weighted between-cluster sum of squares. However, the global $L_1$ penalties in existing SHC frameworks often fail to capture the heterogeneous features that drive data organization at different hierarchical scales. While extensions for cluster-specific feature importance exist, they remain computationally rigid and struggle to account for the non-linear local contexts inherent in high-dimensional data.
In this talk, I will first provide an overview of the transformer attention mechanism and in-context learning for both language and tabular data. I will then introduce a self-supervised framework that treats hierarchical clustering as an in-context reconstruction task. By designing attention heads mapped to specific dendrogram heights, this approach enables scale-dependent and cluster-specific feature selection. I will conclude by presenting preliminary results on tabular datasets and discussing ongoing experimental studies for high-dimensional settings.

Bio: Pratheepa Jeganathan is an Assistant Professor in the Department of Mathematics and Statistics and an Associate Member of the School of Computational Science and Engineering at McMaster University. Her research focuses on multi-view learning through the lens of dependence modeling, developing latent variable and spatio-temporal methods for complex dependent data. Her research lab combines statistical theory, scalable computation, and machine learning to build interpretable, reproducible, and well-validated tools, with applications in spatial omics, longitudinal microbiome data, transportation sensors and GPS data, and loss reserving using multiple loss triangles. Her research is supported by the NSERC and the Canadian Statistical Sciences Institute (CANSSI) Collaborative Research Team Project.

November 26, 2025 - Leveraging LLMs and VLMs in Diagnosis and Treatment.

Speaker: Eman Rezk
When: November 26th, 2025
Time: 12:30 -13:20pm
Where: Hamilton Hall, 410

Abstract: This talk provides an overview of the evolution of text processing, from early text mining methods to large language models and vision language models. We discuss how large language models are finetuned and utilized in medical question answering, diagnosis, and treatment. The presentation will also explain breakthrough vision-language model architectures, their training, and applications. The talk concludes with an overview of our recent work on developing clinically aligned AI systems for dermatology. We present our multimodal pipelines that integrate clinical images with textual lesion descriptions for diagnosis and treatment recommendations.

Bio: Dr. Eman Rezk is an NSERC Postdoctoral Fellow at the University of Waterloo, where her research focuses on developing responsible and equitable artificial intelligence (AI) for healthcare applications. Her work bridges computer vision, machine learning, and medical imaging, with a particular emphasis on fairness across diverse populations. Dr. Rezk received her PhD from McMaster University, where she received multiple national awards, including the Queen Elizabeth Scholarship and the Royal Society of Canada’s Alice Wilson Award, recognizing her contributions to advancing inclusive and interpretable medical AI.

November 12, 2025 - Machine learning methods for sensitivity analysis of climate-economic models

Machine learning methods for sensitivity analysis of climate-economic models.
Speaker: Daniel Presta
When: November 12th, 2025
Time: 12:30 -13:20pm
Where: Hamilton Hall, 410

Abstract: Large scale integrated climate-economic models typically involve numerous underlying parameters with different amounts of uncertainty, some arising from econometric estimates using historical data, some arising from experimental measurement of physical and atmospheric relationships. Because of the intrinsic nonlinear nature of these models, it is often impossible to employ traditional methods for local sensitivity analysis based on comparative statics to understand the effect of a particular parameter on the outcome of the model. Conversely, the high computational cost of the models makes it impractical to use simulation-based methods for global sensitivity analysis. Accordingly, we explore the use of machine learning methods to understand and quantify the influence of multiple parameters, taking full account of nonlinearities, while still being computationally feasible. We illustrate the techniques in the context of an existing stock-flow consistent climate-economic model.

Bio: Daniel Presta is a fourth year PhD Candidate at McMaster University, researching climate-economic modelling under the supervision of Dr. Matheus Grasselli.