Brain-Agnostic 3DCNNs Learn Naturalistic Emotion from 7t fMRI

Abstract

Understanding emotions through neural activity is a key challenge in affective computing and neuroscience. In this work, we leverage brain-agnostic 3D convolutional neural networks (3DCNN) to learn functional representations of emotions from large-scale naturalistic 7T fMRI data. Our learned representations are consistent with neurobiological principles, highlighting the potential of deep learning for neural emotion inference. Code and data are available at [https://github.com/lungerjo/DeepEmotion].