Currently, many biologists must annotate large quantities of data before deep learning can be applied to tasks, and the extra burden in finding objects decreases the likelihood an image can be used for biological discovery. Fortunately, techniques such as few-shot learning can greatly reduce the amount of annotated data needed to train a network. Although these techniques are well-researched for deep learning classification tasks, their application to segmentation or detection has been less explored. My research aims to generalize few-shot learning strategies to new classes of architectures, transforming them from classification tools to segmentation or detection tools. Once implemented, the degree of performance improvement achieved can be evaluated using real biological image data.

Through SuperUROP, I hope to gain authentic research experience that I can use to better inform my career decisions. I’m grateful to have such incredible mentors in the fields of machine learning and biology, and I’m excited to have the opportunity to make meaningful scientific contributions.