Individual Differences in Policy Precision: Links to Suicidal Ideation and Default Mode Network Dynamics

Behavior modeling of decision-making processes has deepened our understanding of behavioral impairments in various psychiatric states. However, as research increasingly emphasizes building models to best explain observed decision-making behavior, it may overlook whether these decisions stem from biologically plausible brain functions. To address this, we developed a probabilistic two-armed bandit task model based on the active inference framework. This model was compared to established reinforcement learning (RL) models and demonstrated superior explanatory power in capturing individual choice variability. A key parameter in our model, policy precision, corresponding to the temperature parameter in the RL model, is also optimized based on previous outcomes. This process effectively explains the balance between model-free and model-based strategies. Incorporating the rate of change in policy precision enhanced the model's ability to explain multi-level brain signals and their inter-subject correlations. Notably, this signal positively correlates with the activation of the default mode network while negatively correlating with the dorsal attentional and frontoparietal networks. Lastly, we validated the relevance of behavioral parameters in a population including individuals with major depressive disorder.