A recorded presentation of my work with butterflies
I recently gave a presentation at the McGuire Center for Lepidoptera Research on my work here at the museum. In the first half I give an introduction to AI-based classification models, the difficulties these models have with biological datasets, and how we designed our BioEncoder package to address some of these issues.
In the second half, I talk about how image encoders can be used to compare complex visual features of diverse groups: I show how we use a pretrained image encoder to generate embedding vectors that capture visual patterns in the wings of Nymphalid butterflies, and how these embeddings then can be used to study the evolution of the color signals themselves.