We conducted several analyses to search for perceptual learning-related changes in early visual representations of stimulus orientation. None of these provided any evidence for perceptual learning. This is in line with findings that monkeys trained on similar visual tasks show only little (Schoups et al., 2001) if any change in the early visual cortex

(Crist et al., 2001 and Ghose et al., 2002). Nevertheless, our combination of fMRI, multivariate decoding, and computational modeling might not be sensitive enough to find any potentially subtle changes in early visual representations. However, our method is sensitive enough to decode the stimulus orientation itself in visual cortex. It is also sufficiently sensitive click here to find learning-related changes in medial frontal cortex. This could suggest an alternative account for perceptual learning which involves higher cortical representations of decision variables. Importantly, this account is in line with results from monkey electrophysiology (Law and Gold, 2008) as well as with recent psychophysical and modeling work (Zhang and Li, 2010, Zhang et al., 2010a and Zhang et al., 2010b). Furthermore, studies investigating

perceptual decisions revealed a similar dissociation between early sensory regions this website and frontal areas (Heekeren et al., 2008 and Romo and Salinas, 2003). Specifically, not sensory areas have been shown to track the physical stimulus properties, whereas neural activity in frontal cortex tracks perceptual judgments and thus the subjective experience of the stimulus (de Lafuente and Romo, 2005, de Lafuente and Romo, 2006, Heekeren et al., 2004, Hernández et al., 2010, Lemus et al., 2010 and Salinas et al., 2000). Our model suggests that reinforcement processes account for perceptual learning. This is in line with recent conceptual work that proposes a common mechanism for perceptual

and reward-based decisions (Rushworth et al., 2009). It is also consistent with recent models of perceptual learning (Seitz and Watanabe, 2005 and Seitz and Dinse, 2007) in which reinforcement signals drive perceptual learning, even if features are task-irrelevant, unattended (Dinse et al., 2003 and Seitz and Watanabe, 2003), or invisible (Seitz et al., 2009). Moreover, besides the behavioral fit of our model, we show that prediction errors correlate with activity in reward-related regions such as the ventral striatum but also in the ACC where perceptual learning-related changes in DV were identified. The presence of activity that correlates with signed prediction errors, the teaching signal in reinforcement learning models ( Kahnt et al., 2009, McClure et al., 2003, O’Doherty et al., 2003 and Pessiglione et al., 2006), provides further evidence for a reinforcement process in perceptual learning.