Fig. 1
From: Dissociating frontoparietal brain networks with neuroadaptive Bayesian optimization
Overview of methodology. a In Experiment 1, a 2D-task space was designed with each dimension corresponding to the probability of 16 tasks recruiting the dFPN or vFPN according to a previous meta-analysis26. Color-coding indicates the hypothesized dissociation (based on this meta-analysis) between the two FPNs: red indicates tasks to be optimal for the contrast dFPN > vFPN (i.e., Wisconsin Card Sorting and Counting/Calculation tasks), while blue indicates tasks to be optimal for the reverse contrast vFPN > dFPN (i.e., Posner, Anti-Aaccade and Go/No-go tasks). This task space was then searched in real-time by the neuroadaptive Bayesian optimization to find optimal tasks that dissociate the dFPN from the vFPN. b In Experiment 2, task parameters of optimal tasks from Experiment 2 were fine-tuned. For the Deductive Reasoning task, the optimization algorithm searched across 1D-space with 16 × 1 difficulty levels. For the Tower of London task, the optimization algorithm searched across 2D-space with 8 (number of steps) x 2 (convolution) parameters. In Experiment 3 (not shown), the same 2D-task space as in Experiment 1 was searched through by the optimization algorithm with the aim of finding task that maximally dissociate the dFPN and vFPN from three other FPNs (for details see main text)
