Machine learning

Basis The architecture we are using is outlined in this paper. Originally, we were planning on using a GAN-based model. But, as outlined in the paper, diffusion-based models have proven to be more clinically relevant. Brownian Bridge Diffusion Model Diffusion consists of two processes, the forward and backwards processes. The forward process consists of applying noise and gradually destroying information of the image, and the backwards process is the removal of noise from an image until our synthetic image is all that remains. So when we want to perform an inference, we give the model pure noise and it does denoising until an image is present. This video does a fantastic job of understanding this process. Specifically, we use a conditional diffusion model, meaning that we give it a starting point, the model applies noise, then denoises; in contrast to just denoising from pure noise. Information about denoising is traditionally learned using a U-net backbone. ...

Dream Team Engineering Dream Team Engineering is a student organization at UF that works closely with Shands hospital at UF. We have teams that build engineering solutions for problems that hospital faculty have. From creating 3D models to help teach the young children receiving operations to better understand their procedure to making apps for patients to use. I joined the DTE in the fall of 2024, first on a team called Design Suture Pads, which was a team designing synthetic skin to be used as a teaching tool for medical students. However, a team that truly piqued my interest was called Deep Brain Stimulation, lead by Hannah Kennedy in collaboration with the Norman Fixel Institute. I applied to switch to this team and joined them in the spring of 2025. ...