This medial frontal brain region is also critically involved in reward-based learning and decision-making ( Behrens et al., 2007, Hayden et al., 2009, Holroyd and Coles, 2002, Ito et al., 2003, Kennerley et al., 2006, Matsumoto et al., 2007 and Rushworth

et al., 2007). Thus, our results suggest that perceptual as well as reward-based learning and decision-making share a common neurobiological basis and that both can be studied in the framework of reinforcement learning. Our results were achieved by combining computational models of reinforcement Venetoclax chemical structure learning with multivariate data analysis methods. Rather than searching for univariate representations of model variables as in conventional model-based fMRI (O’Doherty et al., 2007),

we searched for multivariate representations by using pattern recognition techniques (Haynes and Rees, 2006, Kriegeskorte et al., 2006 and Norman et al., 2006). Multivariate approaches have proven to be more sensitive than univariate approaches for revealing the distributed cortical patterns encoding sensory variables, such as stimulus orientation, motion direction, or color, which are known to be encoded in the joint activity of differentially tuned neurons (Brouwer and Heeger, 2009, Haynes and Rees, 2005, Kamitani and Tong, INCB018424 2005 and Seymour et al., 2009). These patterns have been hypothesized to reflect biased sampling of neural activity (Haynes and Rees, 2005 and Kamitani and Tong, 2005), complex spatiotemporal dynamics involving the vascular system (Kriegeskorte et al., heptaminol 2010 and Shmuel et al., 2010), or large-scale biases (Mannion et al., 2010 and Sasaki et al., 2006). Moreover, recent studies suggest that cognitive and decision variables also are encoded in distributed cortical activity patterns (Hampton and O’Doherty, 2007, Haynes et al., 2007, Kahnt et al., 2010, Kahnt et al., 2011 and Soon et al., 2008).

Taken together, our current approach of decoding variables derived from computational models could provide a fruitful tool to study neurocomputational processes underlying learning and decision-making. In conclusion, here we have shown that behavioral improvements in an orientation discrimination task are accompanied by activity changes in the ACC. Thus, our data provide strong evidence for perceptual learning-related changes in higher order areas. Furthermore, perceptual improvements were well explained by a reinforcement learning model in which learning leads to an enhanced readout of sensory information, which in turn leads to noise-robust representations of decision variables. This learning process involves an updating mechanism based on signed prediction errors, just like classical reward learning. Taken together, these findings support the notion that perceptual learning relies on reinforcement processes and that it engages the same neural processes as reward-based learning and decision-making.