An integrated theory of category-selective regions: evidence from deep neural networks
During my internship with Dr. Talia Konkle at Harvard in Summer 2018, I began a project on the representational structure of category-preferring regions in visual cortex. I tested competing theories of category information in the ventral visual pathway using a set of deep neural networks (DNNs) trained to perform face, scene, and object categorization. Comparing their representational match with human FFA and PPA led us to propose a theory that reconciles the modular and distributed hypotheses of information processing in the ventral stream: that these neural ROIs reflect different facets of a shared feature space supporting a wide range of downstream tasks.
Prince, JS., Konkle, T., (2020). An integrated theory of category-selective regions: evidence from deep neural networks. (Paper in prep.)