Yogatheesan Varatharajah
University of Minnesota, MN·Computer Science·Early-career
Neurologists spend hours visually reading electroencephalograms (EEGs), long multichannel recordings of the brain's electrical activity. This work adds to physician burnout and can lead to errors.
- Fields
Computer Science
Biomedical Engineering
Neuroscience
- Topics
electroencephalography
machine learning
trustworthy AI
- Grants
- 1 known grant, through Aug 2029
What they work on
- Develop an attention-based EEG foundation model, shaped by knowledge of brain signals, that captures how activity changes across recording channels and over time.
- Design self-supervised and weakly supervised training tasks so EEG models can learn from large datasets without relying on scarce, inconsistent expert labels.
- Build model diagnostics and adversarially robust training to detect and correct EEG model failures when patient data differ from the training data.
- Validate the framework on epilepsy subtype classification and treatment outcome prediction with Mayo Clinic and Cleveland Clinic experts, including out-of-sample testing.
More about the research
Machine learning (ML) could ease the load, but most EEG models are trained on human labels that are often inconsistent. They have also rarely been tested on data unlike their training data, so they can fail once used in clinics. The question here is how to build EEG machine learning that clinicians can trust.
Why it matters
Reliable machine learning tools could help neurologists read EEGs faster and more accurately. That could reduce physician workload and errors in diagnosis. Patients with epilepsy may benefit through better subtype classification and predictions of how they will respond to treatment.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Surbhi Goel
University of Pennsylvania, PA·Computer Science·Early-career
Characterize why transformers learn brittle shortcuts rather than robust algorithms, using controlled mathematical environments that isolate real-world reasoning failures and remain tractable for rigorous analysis.
Vishnu Boddeti
Michigan State University, MI·Computer Science
Derive theoretical limits on the trade-offs among predictive accuracy, robustness across patient populations and privacy in healthcare AI models, and evaluate them in practice.
Sanjay Purushotham
University of Maryland-Baltimore County, MD·Computer Science·Early-career
Develop multi-view federated learning algorithms that train shared models across hospitals holding different kinds of patient data, without moving the data between sites.
The school
University of Minnesota: 88 labs on PI Pages, $1.4B a year on research, 722 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.