I am a doctoral student in the School of Computer Science at Carnegie Mellon University (CMU).

My research focuses on advancing time series forecasting for real-world applications by improving machine learning model architectures for enhanced reasoning capabilities. I am currently working on research that explores reasoning in neural networks and techniques that guide model reasoning for conceptual synthesis—directing how models process and combine concepts in predictions without prior exposure to specific data. Guiding model reasoning can enhance the learning of temporal dynamics, bridge data gaps, and mitigate data memorization to improve the accuracy and interpretability of forecasting in zero-shot tasks, particularly for real-world out-of-distribution scenarios.

I am also interested in methods to improve modeling of complex dynamical systems with Deep Learning, such as the intersection of mechanistic modeling and Deep Learning that incorporates first principles from domains in models. This approach can not only enhance model performance but can also facilitate learning underlying mechanics of the system.

I have worked on several projects to improve outcome prediction using forecasting and survival analysis frameworks while contributing open-source code. I have also collaborated with industry partners in healthcare and renewable wind energy to enhance models for critical event prediction.

Research

Time Series Forecasting

Forecasting plays a critical role in decision-making across various domains, from healthcare to finance, by enabling accurate predictions of future events based on historical data. Forecasting models face challenges when encountering out-of-distribution (OOD) data, such as complex compositions of concepts and increasing temporal parameters associated with multiplicative patterns and exponential trends. Research focused on advancing forecasting methods to improve generalization is essential for enhancing reliability and practical application.

Reasoning in Neural Networks

Foundation models are trained on large, diverse datasets, raising a critical question for time series forecasting: do models generalize well because they learn underlying concepts of temporal dynamics, or do they simply memorize specific patterns seen during training? If those models rely on memorization, particularly in the form of time series pattern matching, it could lead to redundancy in the stored knowledge, parameter inefficiency, and possibly limit their ability to generalize well to out-of-distribution (OOD) data. Encouraging these models to rely on reasoning capabilities can mitigate the undesirable memorization of training data and improve model generalization.

Physics-Informed Deep Learning \ Mechanistic Modeling

Efforts to supplement deep learning models with first principles from domain area have demonstrated successful application in many domains such as pharmacology, biology, physics, and fluid dynamics, among others. The combination of domain-specific first principles with deep learning results in models that are not only data-driven but also consistent with the underlying physical, biological, or chemical processes. In this case, deep learning paradigms learn parameters of the system while in turn being constrained by the system design to reflect only realistic modeling solutions.

Publications and Presentations

A complete list of publications can be found here.

Implicit Reasoning in Deep Time Series Forecasting

W. Potosnak, C. Challu, M. Goswami, M. Wiliński, N. Żukowska, A. Dubrawski

This work takes an initial step toward assessing the reasoning abilities of deep time series forecasting models. We find that certain linear, MLP-based, and patch-based Transformer models generalize effectively in systematically orchestrated out-of-distribution scenarios, suggesting underexplored reasoning capabilities beyond simple pattern memorization. To be presented at the NeurIPS 2024 Workshop on Time Series in the Age of Large Models (TSALM).

[Paper] 

Severe Wind Event Prediction with Multivariate Physics-Informed Deep Learning

W. Potosnak, C. Challu, K. G. Oliveras, J. K. Miller, A. Dubrawski

Wind turbines combat climate change by generating clean energy, but their long-term effectiveness depends on minimizing maintenance costs due to damage from severe weather events, such as wind gusts. To address this, we propose a physics-informed deep learning model to better predict severe wind events and a multivariate time series extension. Presented at the Twelfth International Conference on Learning Representations (ICLR).

[Paper]  [Poster]  [Recorded Talk] 

Intraoperative Features Improve Model Risk Predictions Following Coronary Artery Bypass Grafting

W. Potosnak, C. Nagpal, K. A. Dufendach, D. J. Kaczorowski, P. Yoon, J. Bonatti, J. K. Miller, A. Dubrawski

Clinical risk assessment before cardiac surgery guides treatment decisions to reduce postoperative complications. This study aimed to evaluate if machine learning models trained on continuous parameters from intraoperative data can enhance risk prediction compared to the standard STS Risk Calculator for multiple CABG outcomes. Published in the Annals of Thoracic Surgery Short Reports.

[Paper] 

Forecasting Response to Treatment with Global Deep Learning and Patient-Specific Pharmacokinetic Priors

W. Potosnak, C. Challu, K. G. Oliveras, A. Dubrawski

We propose a novel hybrid global-local architecture and a pharmacokinetic encoder that informs deep learning models of patient-specific treatment effects. In the Machine Learning for Health (ML4H) Findings Track Collection.

[Paper]  [Code] 

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] 

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]