Predictive Modeling of Brain-Body Association

Minkyu Choi, Xiaokai Wang, Zhongming Liu

Keywords: foundation model, brain-body association, self-supervised learning

TL;DR: We take a whole-person perspective to model brain-body association through predictive learning and modeling.

Abstract: The brain senses and regulates internal organs through a process called interoception. Appealing theoretical frameworks have been proposed to describe this dynamic interaction, yet they are not supported by computational models that can be validated with experimental data. In this paper, we develop a foundation model to substantiate the embodied predictive interoception coding theory. Through self-supervised learning, the model learns the dynamics of the physiological data and captures their association with brain activities. The model reveals a bilateral and distributed brain network across subcortical to cortical levels that encodes respiratory and cardiac dynamics by predicting their states forward in seconds to tens of seconds. As an initial attempt, our work merits future studies in modeling interoception through concurrent brain and body measurements.

Submission Number: 23