REU Site: Deep Learning for Dynamical Systems in Biological and Physical Sciences

At this REU site, undergraduate student researchers will collaborate closely with faculty mentors in the realm of Deep Learning and Applied Mathematics. Eight NSF-funded students join annually. Participants will engage in an intensive 8-week summer session from May 31 to July 25, 2026. To begin, participants will undergo a short course introducing the structures and coding of neural networks, as well as  their applications in dynamical systems of biological and physical models. Regular daily meetings with mentors will be scheduled to chart out the day's activities, and periodic team gatherings will encourage a consistent pace of advancement. Regular progress updates will be conveyed through weekly presentations. The culmination of your efforts will be showcased through a comprehensive final presentation, either in an oral or poster format, resembling a conference-style setup.

Benefits:

• $5,600 stipend ($700 per week).

• On-Campus housing provided.

• Opportunity for travel awards (up to $400) to present summer research at a national conference.

• Free access to high-end GPUs (including NVIDIA’s A100, H100, and H200).

There are two main objectives: (1) Elevating Deep Learning and Applied Mathematics Research – This objective emphasizes fostering deep learning and applied mathematics research at the undergraduate level. (2) Broadening Research Engagement – This goal aims to increase participation in deep learning and applied mathematics research by integrating interdisciplinary projects that highlight collaborative teamwork and effective problem-solving techniques. 

The training component of this REU Sites program will be anchored in a Python bootcamp, coupled with hands-on workshops in various neural network architectures (e.g. feedforward, convolutional, recurrent, and transformers) and mathematical modeling. Participants will apply these techniques to model complex systems, such as infectious disease spread, neural signal dynamics, ion channel behavior and celestial mechanics. Research combining deep learning with dynamical systems involves designing neural network architectures, solving nonlinear ODE systems, collecting necessary data, and training models.