Sanjay Purushotham
University of Maryland-Baltimore County, MD·Computer Science·Early-career
Machine learning could help doctors personalize treatment and make better decisions, but it needs large amounts of health data. That data is scattered across hospitals, providers and agencies.
- Fields
Computer Science
Biomedical Engineering
Public Health & Epidemiology
- Topics
federated learning
machine learning
computational healthcare
- Grants
- 1 known grant, through May 2028
What they work on
- Develop multi-view federated learning algorithms that train shared models across hospitals holding different kinds of patient data, without moving the data between sites.
- Create coded federated learning methods that use built-in redundancy to cope with sites that have unequal computing resources.
- Design federated adversarial training and verifiable federated unlearning, which provably removes a patient's data from a trained model on request, so models resist attacks and honor deletion requests.
- Generate and evaluate realistic synthetic health data for federated learning, and release an open-source toolkit to test the fairness, interpretability and reliability of models.
More about the research
Pooling it is costly, risky for patient privacy and often blocked by regulations. Federated learning offers a workaround: each site trains a shared model on its own data and shares only model updates, not patient records. In healthcare, however, this approach is fragile. Patient data differ from site to site, sites have unequal computing power, data on some diseases are scarce, and models can be attacked.
Why it matters
Hospitals and health agencies could use these methods to build shared predictive models without exchanging patient records. This would protect privacy while drawing on more diverse data. More robust, fair and attack-resistant models, together with an open evaluation toolkit, could build the trust clinicians and patients need before relying on such tools for clinical decisions and precision medicine.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Lili Su
Northeastern University, MA·Computer Science·Early-career
Build theory for federated learning that uses the statistical structure of each device's data, to explain why it succeeds in practice where optimization-only analyses predict failure.
Mei Liu
University of Florida, FL·Medicine & Clinical Research
Develop methods to build the best individualized cohort of similar past patients from EHR data, as the training set for a personalized AKI risk model for each new hospital patient.
Jie Xu
University of Florida, FL·Computer Science·Early-career
Develop transformer-based vision-language models that outline kidney masses on CT scans, extract clinical concepts from notes, and generate radiology reports, validated against expert annotations.
The school
University of Maryland-Baltimore County: 7 labs on PI Pages, $151M a year on research, 90 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.