The form of this general update equation is reminiscent of RL models. Specifically, the precision-weighting can be understood as (component of) a dynamic learning rate (cf. Preuschoff and Bossaerts, 2007); see Mathys et al. (2011) and section A of the Supplemental

Experimental Procedures for details. In our three-level HGF, two precision-weighted PEs εi occur. At the second level, ε2 is the precision-weighted www.selleckchem.com/products/LY294002.html PE about visual stimulus outcome that serves to update the estimate of x2 (the cue-outcome contingency in logit space). At the third level, ε3 is the precision-weighted PE about cue-outcome contingency that is proportional to the update of x3 (environmental log-volatility). These are the two quantities of

interest that the fMRI analyses in this article focus on. For the exact equations, see the Supplemental Experimental Procedures, section A. The experiment was conducted on a 3T Philips Achieva MR Scanner at the SNS Lab, using an eight channel SENSE head-coil. Structural images were acquired using a T1-weighted sequence. For functional imaging, 500 whole-brain images were acquired in the first fMRI study and 550 images in the second fMRI study, using a T2∗-weighted echo-planar imaging sequence www.selleckchem.com/products/crenolanib-cp-868596.html that had been optimized for brain stem imaging (slice thickness: 3 mm; in-plane resolution: 2 × 2 mm; interslice gap: 0.6 mm; ascending

continuous in-plane acquisition; TR = 2,500 ms; TE = 36 ms; flip angle = 90°; field of view = 192 × 192 × 118 mm; SENSE factor = 2; EPI factor = 51). In order to reduce field inhomogeneities a second order pencil-beam volume shim (provided by Philips) was applied during the functional acquisition. Functional data acquisition lasted ∼21 min. During fMRI data acquisition, respiratory and cardiac activity was acquired using a breathing belt and an electrocardiogram, respectively. fMRI data were analyzed using statistical parametric mapping (SPM8). Following motion correction of Rutecarpine the functional images and coregistration to the structural image, we warped both functional and structural images to MNI space using the “New Segment” toolbox in SPM; see Appendix A in Ashburner and Friston (2005). The functional images were smoothed applying a 6 mm full-width at half maximum Gaussian kernel and resampled to 1.5 mm isotropic resolution. In order to optimize signal-to-noise ratio for critical regions such as the brain stem, we corrected for physiological noise using RETROICOR (Glover et al., 2000) based on an in-house implementation (Kasper et al., 2009) (open source code available at http://www.translationalneuromodeling.org/tapas). For fMRI data analysis, we specified a voxel-wise general linear model (GLM) for each participant.