Mohammad Mahdi Kazemi
PhD Candidate in Computer Science at UBC
Designing robust learning algorithms for reliable medical AI. My research focuses on uncertainty estimation, distribution shifts, and model robustness in clinical settings.
Featured Projects
Selected research and development work.
DEUE: Delta Ensemble Uncertainty Estimation
A novel method for uncertainty estimation in medical imaging AI systems, enabling more reliable predictions of cardiac function from echocardiograms. Published at MICCAI 2022, this work addresses the critical need for AI systems to quantify their own confidence in medical applications.
Lightweight Deep Video Networks for Mobile Devices
Development of efficient deep learning models for cardiac function assessment on mobile devices, enabling point-of-care diagnostics in resource-limited settings.
Latest Thoughts
Writing on AI, research, and technology.
Tail-Risk Control for Vision Systems: Our CVPR 2026 Paper on BQ-SRC
A first look at BQ-SRC, our new CVPR 2026 paper on distribution-free tail-risk control for conformal prediction in modern vision systems.
DEUE: Delta Ensemble Uncertainty Estimation for Robust EF Assessment
A memory and time-efficient epistemic uncertainty estimator for deep regression models in medical imaging, specifically for Ejection Fraction estimation from echocardiography.
Selected Publications
View All Publications→Spectral Conformal Risk Control: Distribution-Free Tail Guarantees via Bayesian Quadrature
CVPR2026
DEUE: Delta Ensemble Uncertainty Estimation for a More Robust Estimation of Ejection Fraction
MICCAI2022
Towards Targeted Ultrasound-guided Prostate Biopsy by Incorporating Model and Label Uncertainty in Cancer Detection
IJCARS2022