Rule activation and ventromedial prefrontal engagement support accurate stopping in self-paced learning

Abstract

When weighing evidence for a decision, individuals are continually faced with the choice of whether to gather more information or act on what has already been learned. The present experiment employed a self-paced category learning task and fMRI to examine the neural mechanisms underlying stopping of information search and how they contribute to choice accuracy. Participants learned to classify triads of face, object, and scene cues into one of two categories using a rule based on one of the stimulus dimensions. After each trial, participants were given the option to explicitly solve the rule or continue learning. Representational similarity analysis (RSA) was used to examine activation of rule-relevant information on trials leading up to a decision to solve the rule. We found that activation of rule-relevant information increased leading up to participants' stopping decisions. Stopping was associated with widespread activation that included medial prefrontal cortex and visual association areas. Engagement of ventromedial prefrontal cortex (vmPFC) was associated with accurate stopping, and activation in this region was functionally coupled with signal in dorsolateral prefrontal cortex (dlPFC). Results suggest that activating rule information when deciding whether to stop an information search increases choice accuracy, and that the response profile of vmPFC during such decisions may provide an index of effective learning.

Model-based fMRI reveals dissimilarity processes underlying base rate neglect

“…As anticipated, a linear mixed effects model collapsed across trial type revealed that pattern similarity to the visual category was the strongest for perfectly predictive features (M = .065), followed by the non-predictive but present features (M = −0.050) and the non-present features (M = −0.145), ( F (2, 42)=54.8, p<0.001). This finding, whereby activation patterns elicited for stimuli during learning are most similar to predictive features, is consistent with recent studies using MVPA to measure dimensional selective attention in categorization and reinforcement learning ( Mack et al, 2013 , Mack et al, 2016 ; Leong et al, 2017 ; O'Bryan et al, 2018 ). For common trials, pairwise comparisons revealed significant differences between pattern similarity to perfect and imperfect predictors ( t (21)=3.38, p=0.003), perfect predictors and non-present features ( t (21)=5.71, p<0.001), and between imperfect predictors and non-present features ( t (21)=4.27, p<0.001).…”

“…These results are consistent with findings from other model-based fMRI studies suggesting that the MTL is involved in similarity-based retrieval ( Davis et al, 2012a , Davis et al, 2012b ). Likewise, the engagement of vmPFC corroborates recent studies suggesting that this region tracks higher relative evidence for categorization decisions ( Davis et al, 2017 ; O'Bryan et al, 2018 ). The positive relationship between vmPFC and similarity processes may also be reflective of attention to strong predictors ( Sharpe and Killcross, 2015 ; Nasser et al, 2017 ) or the application of familiar category rules ( Boettiger and D'Esposito, 2005 ; Liu et al, 2015 ), both of which are consistent with similarity-based accounts of IBRE that attribute choice to a well-established association between perfect predictors and their outcomes that is driven by attention (e.g.…”

“… Mack et al, 2013 ). Consistent with past research using MVPA to decode learned selective attention ( Mack et al, 2013 , Mack et al, 2016 ; Leong et al, 2017 ; O'Bryan et al, 2018 ), multivoxel patterns associated with predictive features were more strongly activated than imperfectly predictive features during the learning phase. Using the same approach to decode which information participants were focusing on during ambiguous test trials, we found stronger activation of patterns associated with common compared with rare stimulus features, but importantly, this pattern only emerged in cases where participants chose the rare category.…”

“…We predict that our multivoxel results are not driven directly by simple feature-based attention, but instead indicate some combination of attention and memory-based retrieval of the category exemplars. Pattern similarity measures in ventral temporal cortex have been shown to effectively index both dimensional selective attention ( Leong et al, 2017 ; O'Bryan et al, 2018 ) and the retrieval of non-present, associated stimuli (e.g. Zeithamova et al, 2012 ; for review, see Rissman and Wagner, 2012 ).…”

Occipitotemporal Representations Reflect Individual Differences in Conceptual Knowledge

“…Previous studies have relied on contrastive analyses in which neural representations of attended stimulus dimensions are compared to those of unattended dimensions. Although statistically-powerful, this approach defines selective attention in terms of the experimental paradigm (but see O’Bryan et al, 2018), and therefore sidesteps effects associated with individual differences in conceptual knowledge (e.g., Craig & Lewandowsky, 2012; Little & McDaniel, 2015; McDaniel et al, 2014; Raijmakers et al, 2014). These effects can be substantial, particularly for ill-defined categorization-problems (such as the 5/4 categorization task) that are common in every-day life (Hedge et al, 2017; Johansen & Palmeri, 2002).…”