Nasim of Marquette University, Milwaukee.

Whole slide images (WSIs) provide rich spatial and contextual information for cancer diagnosis,
yet their gigapixel scale and susceptibility to artifacts, adversarial noise, and distribution shifts
complicate reliable deep learning integration. In this talk, a unified graph-based framework for
robust and uncertainty-aware WSI analysis is presented. The proposed approach represents WSIs
as graphs to capture spatial dependencies and leverages graph neural networks (GNNs),
including Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and
GraphSAGE, for feature learning. To enhance robustness, a denoising module, enables input-
specific optimization that mitigates both acquisition-related artifacts (e.g., motion, staining
variability) and adversarial perturbations at image and graph levels. A transformer-based module
further refines global contextual modeling for accurate cancer classification across multiple
datasets. To address reliability under distribution shifts, we deploy a multi-head GNN framework
for predictive uncertainty estimation, where head-wise divergence provides a principled measure
of confidence and supports out-of-distribution detection. Extensive evaluations demonstrate that
the proposed framework significantly improves classification accuracy, robustness to noise and
attacks, and uncertainty quantification compared to conventional and non-robust methods.

Biography:

Nasim Yahya Soltani received the Ph.D. degree in electrical engineering from the University of
Minnesota, Minneapolis, MN, USA, in 2014. From 2014 to 2017, she was a Research Associate
with the Digital Technology Center at the University of Minnesota. She subsequently served as a
Senior Data Scientist at Harley-Davidson Motor Company, Milwaukee, WI, USA, from 2018 to
2019. Since 2019, she has been a Northwestern Mutual Assistant Professor in the Department of
Computer Science at Marquette University, Milwaukee. Her research lies at the intersection of
statistical signal processing, machine learning, optimization theory, and network science, with
applications in healthcare, wireless communications, and smart grid systems.
Dr. Yahya Soltani has an established record of scholarly contributions, with publications in
leading peer-reviewed IEEE journals and conferences, reflecting both theoretical innovation and
practical relevance.