Figure 1
The predictive processing account of PLUMM. A) Rhythms with three levels of syncopation lead to meter-based predictions whose uncertainty depend on both the position in the meter and the strength of the metrical model. B) Moderately syncopated rhythms maximize the number of strongly weighted prediction errors. Adapted from Stupacher et al., 2022.
Figure 2
The learning progress hypothesis. Humans are intrinsically motivated for learning progress, which is operationalized as the rate of prediction error minimization over time. The detection of reducible prediction errors mobilizes resources associated with state curiosity to maximally capitalize on the learning potential. Learning progress is registered as pleasure and enhances memory encoding, which in turn facilitates further learning progress, setting up a feedback loop. Adapted from Oudeyer et al., 2016.
Figure 3
A neuroscientific model of the LP account of PLUMM. A) Phasic pulses of nigrostriatal dopamine into the dorsal striatum initiate cycles of meter-based timing mechanisms via excitatory and inhibitory signals within the motor corticostriatal loop. Adapted from Cannon & Patel, 2020. B) The detection of reducible prediction errors relative to the metrical model leads to mesolimbic dopamine signals to the hippocampus to enhance memory, and to the dACC which in turn activates the LC to release norepinephrine, leading to the mobilization of attentional resources. The PFC updates metrical models along with higher level schemas. Adapted from Ripollés et al., 2016 and Silvetti et al., 2018. dACC, dorsal anterior cingulate cortex; Hipp, hippocampus; LC, locus coeruleus; NAc, nucleus accumbens; PFC, prefrontal cortex; SMA, supplementary motor area; SN/VTA, substantia nigra/ventral tegmental area; VP, ventral pallidum.