Eric Zimmermann
Senior Applied Scientist, Biomedical ML
I am a Senior Applied Scientist at Microsoft Research New England. My work focuses on machine learning methods for scientific and biomedical data, with interests in biomedical machine learning, representation learning, and multimodal systems.
Much of my recent research has focused on computational histopathology and fluorescent microscopy at scale. This includes work published in Nature Medicine on clinical grade cancer detection.
Previously, I worked at Sama on data curation and dataset quality, and at CAE as an electrical system designer for full-flight simulators.
I completed my M.Sc. and B.Eng. in Electrical and Computer Engineering at McGill University and Mila, where I worked in the Probabilistic Vision Group under Tal Arbel. My research focused on self-supervised learning, alignment, and representation geometry, with applications to brain MRI analysis for detecting and predicting disease progression in multiple sclerosis.
I am always happy to hear from people interested in related research or potential collaborations; the best way to reach me is ezimmermann at microsoft dot com.
Outside of work, I enjoy hiking, video games, and excessive amounts of coffee.
Research
My research focuses on self-supervised and multimodal learning for scientific and biomedical data. I am especially interested in methods that learn useful representations from large, heterogeneous datasets and transfer across tasks, institutions, scales, and measurement modalities.
Much of my recent work applies these ideas in biomedical settings, where labels are scarce, measurements are heterogeneous, and evaluation must be robust across datasets and deployment contexts. Current research interests:
- Applications of self-supervised learning to biomedical problems, including histopathology, fluorescence microscopy, and single-cell RNA sequencing.
- Understanding the theory and geometry of representation learning.
- Multimodal modeling and integration across heterogeneous data sources.
Selected work
-
KerJEPA: Kernel Discrepancies for Euclidean Self-Supervised Learning
arXiv:2512.19605, 2025.
-
A foundation model for clinical-grade computational pathology and rare cancers detection
Nature Medicine 30(10), 2924-2935, 2024.
-
PRISM2: Unlocking Multi-Modal General Pathology AI with Clinical Dialogue
arXiv:2506.13063, 2025.
-
Virchow2: Scaling Self-Supervised Mixed Magnification Models in Pathology
arXiv:2408.00738, 2024.
-
PRISM: A multi-modal generative foundation model for slide-level histopathology
arXiv:2405.10254, 2024.
Publications
-
Tackling the complexity of cancer with generative models
Cell 189(8), 2218-2231, 2026.
-
arXiv:2603.00193, 2026.
-
arXiv:2602.22176, 2026.
-
Beyond alignment: synergistic integration is required for multimodal cell foundation models
bioRxiv, 2026.02.23.707420, 2026.
-
KerJEPA: Kernel Discrepancies for Euclidean Self-Supervised Learning
arXiv:2512.19605, 2025.
-
PRISM2: Unlocking Multi-Modal General Pathology AI with Clinical Dialogue
arXiv:2506.13063, 2025.
-
Virchow2: Scaling Self-Supervised Mixed Magnification Models in Pathology
arXiv:2408.00738, 2024.
-
PRISM: A multi-modal generative foundation model for slide-level histopathology
arXiv:2405.10254, 2024.
-
Adapting self-supervised learning for computational pathology
DCA in MI Workshop, CVPR 2024; arXiv:2405.01688, 2024.
-
A foundation model for clinical-grade computational pathology and rare cancers detection
Nature Medicine 30(10), 2924-2935, 2024.
-
Benchmarking a benchmark: How reliable is MS-COCO?
DataComp Workshop, ICCV 2023; arXiv:2311.02709, 2023.
-
An empirical study of uncertainty in polygon annotation and the impact of quality assurance
DataComp Workshop, ICCV 2023; arXiv:2311.02707, 2023.
-
Consensus learning with multi-rater labels for segmenting and detecting new lesions
MSSEG-2 challenge proceedings: Multiple sclerosis new lesions segmentation challenge using a data management and processing infrastructure, 2021.