Nate Britton's Mid-Sure Project Presentation (Transfer learning)

In the summer of 2021 Nate Britton worked with me as project titled "Reexamining Transfer Learning Image Segmentation Hypothesis by Scaling Up"

Image segmentation is a process in which an image is separated into foreground (areas of interest) and background regions. Image segmentation is used as a first step in many research fields, which is why the SEE-Insight Team has decided to focus on it as one of its primary scientific image analysis workflows. Building upon years of software developed by the SEE-Insight Team as well as previous work, this research will further explore how transfer learning (using previous results to inform better future ones) can help facilitate the search for segmentation algorithms. This work builds off prior research with transfer learning that used a low number of iterations. This prior work found that there was little to no difference in results between transfer learning and randomized algorithm/parameter selection in image segmentation accuracy. As there is still good reason to expect that transfer learning should yield better-than-random results, the goal of this project is to continue the research by experimenting with finding better baseline algorithms by exploring more iterations and greater population sizes. Research on this project has been and will continue to be implemented using Python and shared as an open-source project on GitHub. In the presentation of this research, I will explore the additional progress that has been made thus far in researching and implementing transfer learning.

Here is a video from his final presentation: