Dynamical Bayesian cue integration in a volatile Posner dual-cueing task

This research investigates the Bayesian integration of visual and auditory cues in a dynamic decision-making task. We introduce a volatile Posner dual-cueing task to simulate the volatility nature of sensory information and develop a Bayesian agent to model human behavior. The agent employs a hierarchical Bayesian model to estimate the validity of visual and auditory cues, which are assumed to follow a multivariate Bernoulli distribution with a time-varying coefficient of variation (CV). The agent's response model integrates the cues' attentional factors, weighted by their precisions, to predict reaction speed (RS). Through simulations, the agent demonstrates the ability to adapt to changes in cue validity and precision, reflecting the dynamic allocation of attention and the optimization of reaction speed. The results show that the agent's performance is consistent with theoretical predictions, suggesting that the model is promising to effectively captures the essence of human multimodal integration. The study paves the way for future work to validate the model with human behavioral data and enhance our understanding of cognitive processes underlying attention and decision-making.