Developing AI and advanced statistical time series models for heartbeat dynamics
Sandya Subramanian, Professor
Computational Precision Health
Applications for Spring 2026 are closed for this project.
Our lab works on developing new AI-based and statistical tools for analyzing various physiological time series. One of the most important among these is heartbeat dynamics, which is the beat-to-beat variation in electrical activity to the heart. We develop models that blend deep neural networks and traditional statistics, such as neural temporal point processes (NTPP), and build upon them, including zero-shot model formulation and ectopic beat correction. We also investigate time series foundation models.
Role: Tasks
1. Benchmark various implementations of the NTPP-based heartbeat dynamics model, including zero-shot, online, and pretrained versions
2. Benchmark ectopic beat correction models that utilize the NTPP structure
3. Investigate pretraining methodologies to pretrain NTPP models on large ECG datasets.
4. Fine-tune various publicly relaeased time series foundation models on our in-house ECG data to measure their performance and compare with NTPPs
Learning Outcomes
1. The student will gain a strong understanding of building and training models with modern machine learning frameworks such as Pytorch.
2. The student will build understanding of time series methodologies from statistical, to deep learning, to LLM-based.
Qualifications: 1. Have taken at least one upper level undergraduate statistics course, with exposure to the mathematical theory of state space models and point processes
2. Have taken several courses with coverage of machine learning and AI concepts, including both mathematical theory and practical implementation
3. 2+ years coding experience in Python, specifically including deep learning frameworks (Pytorch or equivalent). Experience training machine learning models.
4. Have taken at least one basic biology course or have some exposure to biology/medicine through other means
5. Experience working together as part of a team
6. Be highly organized and good at managing time, even during weeks with midterms/other assignments.
Hours: 9-11 hrs
Related website: https://www.subramanianlab.com
Mathematical and Physical Sciences Engineering, Design & Technologies Biological & Health Sciences