Willa Potosnak is a 1st year Ph.D. student at the Carnegie Mellon University (CMU) Robotics Institute (RI). She is currently researching Machine Learning (ML) and statistical methods to improve Forecasting algorithms. Among her research goals are to improve ML methods used to aid intervention and best practice decision making. She is particularly interested in the intersection of Forecasting, State-Space Modeling of Dynamical Systems, and Causal Inference to address clinical challenges within healthcare.


Projects include both research as well as volunteer work.


Time series healthcare data can be challenging to forecast since there are many exogenous factors that contribute to changes in signal trajectories or patterns. Incorporating exogenous information in models can improve forecasts and reduce variance in model predictions. Exogenous information, such as medications, are often represented as sparse variables in data, yet may have latent time-dependent effects not encapsulated in numeric data. I am currently researching Machine Learning approaches to develop and parametrize mathematical models that encapsulate relevant time-dependent information for use in improving forecasts for data with sparse features.

Cardiovascular Surgery Risk Prediction

Preoperative risk prediction algorithms, such as the STS Risk Calculator, are frequently used in practice by clinicans to aid intraoperative and post-operative decision-making. The objective of this research effort is to improve individual patient risk predictions for multiple outcomes, such as mortality and renal failure, following cardiac-surgeries, including coronary artery bypass grafting (CABG). The broader inclusion of time-series intraoperative data in risk models to improve post-operative predictions has been one focus of the research effort.

Federated Learning

Developing robust machine learning algorithms can require large amounts of data. Generally data may be collected from different institutions and shared between multiple servers or devices for this purpose. However, data sharing can be restricted due to data privacy and security constraints. Federated learning is one Machine Learning technique that addresses data sharing constraints by training algorithms across decentralized data servers. Current research includes leveraging Knowledge Distillation algorithms to personalize ML models for institutional organizations with data privacy and communication bandwidth constraints.


RoboticsEd is a website of STEM resources developed with the goal of making robotics education more accessible to educators and their students. I am currently a volunteer co-lead of website design and content for a team of students and educators from the robotics community working on this initiative. This initiative is also co-lead by Dr. Maxim Likhachev, Dr. Oliver Kroemer and Rachel Burcin of the CMU Robotics Institute.


Publications and Presentations

A complete list of publications can be found here.

auton-survival: an Open-Source Package for Regression, Counterfactual Estimation, Evaluation and Phenotyping with Censored Time-to-Event Data

W. Potosnak, C. Nagpal, A. Dubrawski

auton-survival is an open-source repository of tools for survival regression, adjustment in the presence of domain shift, counterfactual estimation, phenotyping for risk stratification as well as estimation of treatment effects with censored time-to-event data. In Proceedings of the 7th Machine Learning for Healthcare Conference, PMLR.

[Paper]  [Blog]  [Code] 

Machine Learning Models with Intraoperative Features Improve Risk Predictions Following CABG

W. Potosnak, K. A. Dufendach, A. Kaczorowski, J. K. Miller, A. Dubrawski

Current models that predict patient risk of adverse outcomes following coronary artery bypass grafting (CABG) rely mainly on preoperative data. Machine Learning was used to determine whether inclusion of intraoperative data improves risk predictions for multiple outcomes following CABG. Presented at the 2022 Society of Thoracic Surgeons Coronary Conference.


Robust Rule Learning for Reliable and Interpretable Insight into Expertise Transfer Opportunities

W. Potosnak, S. Caldas, K. A. Dufendach, J. K. Miller, A. Dubrawski

We propose the use of non-parametric Machine Learning algorithms to improve the reliability of algorithmic rule learning. Rule learning was applied to identify beneficial model transfer opportunities between healthcare sites. In Proceedings of the Thirty-Sixth AAAI Conference on Artificial Intelligence.

[Paper]  [3-minute Video]

Graph Neural Networks for Improved El Nino Forecasting.

S. Rühling Cachay, E. Erickson*, A. F. C., Bucker*, E. Pokropek*, W. Potosnak*, S. Osei, B. Lütjens.

We propose the application of spatiotemporal Graph Neural Networks (GNN) to forecast El Niño–Southern Oscillation (ENSO) at longer lead times with finer granularity and improved predictive skill than current state-of-the-art methods. Presented at the 2020 NeurIPS workshop, Tackling Climate Change with Machine Learning.

[Paper]  [Recorded Talk]


For questions regarding research, please reach out to the following email.