Research interests
Trustworthy machine learning for medical and scientific imaging. I work on making diagnostic models usable outside the conditions they were trained in: efficient vision transformer architectures, calibrated uncertainty and selective prediction, and post-hoc explanation methods. I am equally interested in the evaluation side of this problem: how much of a reported accuracy gain survives correction for fold dependence, and whether aggregate performance conceals disparities across patient subgroups.
Publications
Journal
- Rahman, M. M., Gony, M. N., Ullah, M. S., Shuvra, S. M. K., Haque, R., Ahmed, M. R., et al. (2026). MaizeFormerX: a lightweight vision transformer with cross-scale attention for explainable maize leaf disease diagnosis. Scientific Reports. doi:10.1038/s41598-026-44550-0. First author.
Conference papers: accepted
- Rahman, M. M., Hiba, S., Nahar, S., Anjum, N., Anam, K., Mukherjee, S., Hossain, M. S. Beyond accuracy: unequal errors in machine learning models for Alzheimer’s disease prediction on OASIS and ADNI. 2026 IEEE International Conference on Optics, Machine Learning and Emerging Technology (OMLET), Nairobi, Kenya. Accepted. First author.
- Rahman, M. M., Anjum, N., Hiba, S., Katkoriya, R. N., Rustamova, A., Fahim, S., Lesch, R. Leakage-aware intersectional fairness mitigation for distributed rating classification in recommender systems. 2026 IEEE OMLET, Nairobi, Kenya. Accepted. First author.
Conference papers: published
- Rahman, M. M., Das, N., Uddin, M. S., Shammah, R. S., et al. (2025). An investigation of face detection and feature extraction techniques in deepfake detection using a multi-model machine learning approach. 2025 3rd International Conference on Business Analytics for Technology and Security (ICBATS), British University in Dubai, UAE. doi:10.1109/ICBATS66542.2025.11258249. First author.
- Mohiuddin, A. B., Rahman, M. M., Gony, M. N., Shuvra, S. M. K., Rafy, A., Ahmed, M. R., et al. (2025). Accelerated and accurate cervical cancer diagnosis using a novel stacking ensemble method with explainable AI. 2025 International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN), Saidpur, Bangladesh. doi:10.1109/QPAIN66474.2025.11171850. Co-author (second author).
- Abid, S. M., Xiaoping, Q., Islam, M. M., Islam, M. A., Rahman, M. M., Rafy, A., Alam, M. Z. (2025). Edge-conditioned GAT for journal ranking in citation networks. IEEE 6th International Seminar on Artificial Intelligence, Networking and Information Technology. Co-author (fifth author).
Current research
Stain-aware cross-attention segmentation for breast histopathology (SMCA-HistoSegNet). San Francisco Bay University, 2026 to present. In progress. The project transfers the cross-scale attention idea from MaizeFormerX to segmentation of H&E breast histopathology slides. Stain handling is treated as an experimental variable, comparing no normalization, conventional normalization, stain augmentation, and a learnable stain-aware representation. It studies explanation faithfulness and stability under stain perturbation, and uncertainty with temperature scaling, MC dropout, deep ensembles, and split-conformal deferral, reporting calibration degradation separately from accuracy degradation under cross-site shift.
Links
Google Scholar · GitHub · LinkedIn