In recent years, there have been many computational models exploring how spontaneous neural dynamics arise out of the brain's structural connectome. Most of these models have focused on the resting state (with no external input) or investigate the effects of simple sensory inputs. However, neural systems exist to process incoming information from the environment and to guide motor output. Here, we explore a very simple neural network model (similar to that used previously to model dynamics at rest) and allow it to control a virtual agent in a very simple environment. This setup generates interesting brain-environment interactions that suggest the need for homeostatic mechanisms to maintain rich spontaneous dynamics. We investigate roles for both local homeostatic plasticity as well as macroscopic task negative activity (that compensates for task positive, sensory input) in regulating activity in response to changing environment. Our results suggest complementary functional roles for both balanced local homeostatic plasticity and balanced task-positive task negative activity in maintaining simulated neural dynamics in the face of interactions with the environment. This work highlights the challenges that feedback between environment and brain presents to neural models as well as the different roles that balancing mechanisms may play; it also suggests a functional role for macroscopic task negative systems in the brain that may be of relevance to understanding large-scale brain systems such as the default mode network.
|Publication status||Published - 11 Jun 2016|