Fig. 3.

Mapping between stimuli and neural responses using the Bayesian optimization approach. (a and b) Experiment parameter space estimates for each subject (sub_), taking all available runs into account. The color bar represents the estimates by the Bayesian method on how optimal the experimental condition is for evoking the target brain state: the higher the predicted value, the more optimal the stimuli combination (yellow); the lower the predicted value, the less optimal the stimuli combination (dark blue). The Bayesian optimization accurately recovers the hypothesized relationship between stimuli and neural responses: with optimal stimuli combinations in the center of the grid and least optimal stimuli combinations in each of the grid's corners. To facilitate visual inspection, the exact coordinate of the empirical optimum (i.e., maximum predicted value) is marked as red dashed line. (c) Post-hoc fMRI pattern of activation. The orange dots in (b) were entered into the general linear model (GLM) as observations for the regressor modeling the most optimal stimuli combinations while the dark blue dots entered the GLM as least optimal stimuli combinations. The cluster corrected results of the higher-level (summarized over all runs) analysis are shown. The contrast ‘most optimal stimuli combinations > least optimal stimuli combinations’ is shown in yellow while the contrast ‘least optimal stimuli combinations > most optimal stimuli combinations’ is rendered in dark blue. The data summarize results from all four runs for all individuals except for sub_04 who only completed a single run due to MRI technical failure during scanning.