Encoding the Brain: An Investigation of Models for Improved Neural Activity Predictions of Perceived Images

Abstract

This paper develops various iterations of neural encoding models. My work on these models represents my personal contributions to the collaborative work of MedARC, an open community of ML, medical, and neuroscience researchers supported by Stability AI, hoping to submit state of the art neural encoding predictions to the 2023 Algonauts challenge. The models I helped develop and implement include linearizing encoding models, ensemble models, multi-modal models, and a novel model I call CycleBrain. This work has strong implications for research to understand the brain, medical applications for investigating neurological diseases, and even long term technological applications for brain-computer interfaces.