Evolutionary Generated Parcellations
This project, while ultimately unsuccesful, aimed to generate novel parcellations based on their predictive merit. To do this, a specially designed evolutionary search was performed, where parcellations themselves were mutated and evaluated. The project source code is hosted through github.
A slide deck containing a detailed description of the project, as well as results + interpretation can be found here.
The eventual key take-away from this work was actually the realization that parcellations with higher numbers of parcels performed better than their lower parcel counter-parts. This led to follow-up work in Performance Scaling for Parcellations