Robust & Generalizable ML
Learning methods that account for shifts across sensors, scenes, and data distributions.
Hello, I am
Mazharul Hossain
AI Researcher in Computational Imaging & Scientific Machine Learning
I develop robust, generalizable, physics-informed machine learning methods for scientific and biomedical imaging.
Explore my research
Computational imaging · Robust ML · Scientific AI
My Research
I develop machine learning methods for scientific images, with a focus on reliability when data or measurement conditions change.
Computational Imaging
Reconstruction methods informed by how images are measured and formed.
Biomedical & Scientific AI
Machine learning for scientific and biomedical imaging questions.
Selected research
From measurement challenges to reliable models
GMoE-AD graphical abstract: assets/images/work/gmoe-ad-graphical-abstract.png
Hyperspectral Anomaly Detection Across Sensors
- Problem
- Rare anomalies and changing backgrounds make it hard for any single detector to work reliably across hyperspectral scenes and sensors.
- Method
- HUE-AD combined spectral-abundance estimates with three anomaly detectors by equal vote. GE-AD replaced fixed selection and weights with a learned ensemble; a later study compared supervised and unsupervised variants. GMoE-AD added neural experts and domain-adversarial training for cross-domain detection.
- Evidence
- GE-AD averaged 86.5% F1-macro on eight public images. GMoE-AD averaged 0.943 ROC-AUC across 22 images from seven datasets with multi-domain training and 0.910 in leave-one-dataset-out evaluation.
UPIGAN GAN block diagram: assets/images/work/upigan-gan-block-diagram.png
Physics-Informed Reconstruction for 3D Microscopy
- Problem
- Reconstructing 3D structured-illumination microscopy images from noisy measurements requires both fine detail and fidelity to how the microscope forms an image.
- Method
- USR combines a learned image prior with repeated corrections based on the microscope point spread function. UPIGAN extends this physics-informed, unrolled approach with adversarial learning to improve reconstruction.
- Evidence
- Both studies used 18 fixed-cell mitochondrial samples. Their visual comparisons and reported PSNR and SSIM plots favor physics-informed unrolling over the respective non-physics baselines, but neither paper reports an independent external test.
Map901 project banner: assets/images/work/w-1.png
Map901: Indoor Hazard Mapping for Public Safety
- Problem
- Emergency responders need more than floor plans: they need to know where safety-critical objects are inside complex buildings.
- Method
- The Map901 team combined backpack LiDAR and 360° video, segmented safety objects in images, transferred their labels into 3D point clouds, and stitched and georeferenced the building scans.
- Evidence
- The studies surveyed seven Memphis buildings totaling 1.86 million square feet and addressed 30 classes of public-safety objects. The expanded study used 2,964 labeled images; its 3D maps still required manual cleanup of some labels.
Research output
Selected Publications
Representative work in computational imaging, robust machine learning, and scientific AI.
- 2026
- 2025
Improving Semantic Segmentation through Task Adaptation for UAV Hyperspectral Agricultural Imagery
SPIE Autonomous Air and Ground Sensing Systems
- 2024
Greedy Ensemble Hyperspectral Anomaly Detection
Journal of Imaging
- 2023
- 2022
Building Rich Interior Hazard Maps for Public Safety
Springer Communications in Computer and Information Science
My Expertise
My Recent Projects
About
Building reliable AI for scientific imaging
I am Mazharul Hossain, Ph.D., a Visiting Fellow at the National Cancer Institute, part of the National Institutes of Health. I develop machine-learning and scientific-computing methods for imaging problems where accuracy must hold up beyond the conditions used to build a model. My current work focuses on computational cryo-electron microscopy, while my earlier research explored physics-guided 3D structured-illumination microscopy reconstruction.
During my Ph.D. in Computer Science at the University of Memphis, I developed a body of work in hyperspectral image analysis, including spectral-unmixing, ensemble learning, learned detector selection, supervised and unsupervised anomaly detection, UAV crop segmentation, and cross-domain generalization. This progression culminated in my first-author GMoE-AD paper, published in Sensors in 2026, which combines anomaly detectors, a pretrained hyperspectral model, and a mixture of neural experts to address changes across sensors and datasets.
My master’s work produced a multimodal indoor-mapping solution that combined LiDAR, 360° imagery, and object detection to build public-safety hazard maps. I have also studied explainable AI, with an emphasis on making model requirements, assumptions, and evaluation understandable to the people who use and assess learning systems.
Contact
Let’s Connect
I welcome conversations about computational imaging, scientific machine learning, and research collaboration.