Is a patient with muscle aches, headache and fever more likely to have influenza or Ebola? Most people correctly choose ‘influenza’ because it is the more common of the two diseases. But what about someone with a cough and unexplained bleeding? Coughing is a symptom of influenza but not of Ebola. Unexplained bleeding is a symptom of Ebola but not of influenza. Faced with ambiguous symptoms such as these, many people would diagnose ‘Ebola’ despite knowing that influenza is more common. Indeed, when a situation shares characteristics with both a common and a rare category, we often tend to predict that it belongs to the rare group. This phenomenon is known as base rate neglect, but why does it occur?

One theory is that we pay more attention to features that belong to rare categories (such as unexplained bleeding) because they are distinctive and unusual. But another possibility is that we use our knowledge of the common category to rule out examples that do not conform to our expectations. Of the many cases of influenza that you have heard about or experienced, probably none of them featured unexplained bleeding.

To distinguish between these possibilities, O’Bryan et al. trained healthy volunteers on a categorization task that included ambiguous stimuli. The participants performed the task inside a brain scanner. O’Bryan et al. then programmed a computer to solve the same problems. The simulation either used similarity-based judgments (how similar is this to the rare category?), dissimilarity-based judgments (how dissimilar is this from the common category?), or both. The results suggested that when people show base rate neglect, they rely more on dissimilarity-based evidence than on similarity-based evidence. In other words, they focus more on how a test item differs from the common category. Consistent with this, whenever the volunteers chose the rare category, their brains were processing information about the common category. The imaging results also revealed that when the volunteers used dissimilarity-based evidence, they activated a brain region involved in abstract thinking and reasoning.

How people use information about likelihoods is relevant to all aspects of decision-making. Beyond helping us to understand how we assign items to categories, the work by O’Bryan et al. could also inform future research in areas such as learning and memory.

Does this patient have influenza or Ebola virus? Categorization is a fundamental process that underlies many important decisions. Categories, such as viruses, often have different relative frequencies or base rates. Influenza, for example, is very common and infects millions of people worldwide each year, whereas Ebola virus tends to have infection rates that are orders of magnitude lower.

One critical question is how people use such base rate information when making categorization decisions. Research so far has suggested that people tend to be, at best, inconsistent in their use of base rate information. In both realistic studies with medical professionals and artificial categorization tasks in the lab, when confronted with examples that share characteristics with both rare and common categories, people show a tendency to predict the rare category much more often than the base rates would suggest (Tversky and Kahneman, 1974; Casscells et al., 1978; Bravata, 2000). In an extreme case, known as the inverse base rate effect (IBRE), people may even predict rare categories as more likely than common ones (Medin and Edelson, 1988). For example, in an IBRE context, a patient presenting with cough (a characteristic feature of influenza) and unexplained bleeding (a characteristic feature of Ebola), may be more likely to be diagnosed with Ebola than influenza.

The mechanisms that lead to base rate neglect are currently undetermined at both the cognitive and neural levels. Computationally, according to influential work with similarity-based categorization models (Medin and Edelson, 1988; Kruschke, 1996, Kruschke, 2001), the IBRE arises from differential selective attention to features for common and rare categories. Specifically, participants learn to attend more strongly to features of rare categories, making ambiguous cases seem more similar to rare categories and thus more likely to be rare category members. In terms of the flu example, participants may attend more to the unexplained bleeding feature of the rarer Ebola virus category, and thus predict Ebola when confronted with a patient with both features.

Similarity-based category learning models have strong support in the neurobiological category learning literature. Model-based predictions for how similar items are to stored category representations have been shown to correlate with activation in the medial temporal lobes (MTL; Davis et al., 2012a, Davis et al., 2012b). Moreover, at a finer-grained level, multivoxel activation patterns in the MTL have been shown to contain information associated with higher-order similarity relationships between category members anticipated by similarity-based models (Davis and Poldrack, 2014), including those predicted by differences in selective attention (Mack et al., 2016). The dorsolateral prefrontal cortex (dlPFC) tends to track predictions of choice uncertainty from similarity-based models, whereas ventromedial PFC (vmPFC) tends to track estimates of high choice accuracy or model-confidence (Davis et al., 2017).

Despite the strong cognitive and neural evidence for similarity-based models, it remains an open question whether they provide a complete account of IBRE-like phenomena. One alternative proposition is that people’s choice of rare categories when confronted with conflicting information may stem from reliance on dissimilarity processes, either solely, or in addition to similarity-based processes. According to theories that focus on dissimilarity-based processes, people build strong expectations of the common category; thus they view items containing features inconsistent with these expectations as more likely to be members of the rare category (Juslin et al., 2001; Winman et al., 2005). For example, a doctor may have seen thousands of cases of flu, none with unexplained bleeding, and thus rule out influenza and choose Ebola virus based on these expectations. In these cases, it is dissimilarity to members of the common (unchosen) category that drives choice, rather than the similarity to rare (chosen) category members per se.

Formal models positing dissimilarity processes have so far been explicitly dual-process oriented. For example, ELMO, a computational model that incorporates a choice elimination decision based on dissimilarity, argues that such elimination depends on explicit reasoning processes that are separate from similarity-based processes that arise in other trials (Juslin et al., 2001). In the present study, we propose a new account based on a recently proposed dissimilarity-based extension of the generalized context model, the dissGCM (Stewart and Morin, 2007). This account uses the same basic similarity computations as standard similarity-based models (e.g. Nosofsky, 1986), but allows similarities and dissimilarities to stored exemplars to be used as evidence for a category. In terms of the above example, dissimilarity to influenza can be used as evidence for Ebola (and vice versa).

As specified computationally, the dissGCM is agnostic about whether using dissimilarity-based evidence constitutes a different cognitive or neurobiological mechanism from using similarity-based evidence. On one hand, the dissGCM has no fundamentally different computations from a basic similarity process; as detailed below, dissimilarity is a simple transformation of similarity. On the other hand, it is possible that dissimilarity processes require manipulation of similarity relationships between category representations in a more symbolic or abstract manner, as anticipated by previous dissimilarity theories. Given that the dissGCM makes separable estimates for the relative contributions of similarity- and dissimilarity-based evidence to choice in a given trial, these model predictions can be used to explicitly test whether employing dissimilarity-based evidence against unchosen categories engages brain regions beyond those associated with the use of similarity-based evidence, consistent with dual-process accounts of IBRE. Specifically, we can test whether regions that are known to be critical for using higher-level abstract rules track dissGCM’s predicted trial-by-trial used of dissimilarity-based evidence, and whether these regions diverge from those typically found to track estimates of similarity-based evidence.

Higher-level cognitive control mechanisms are thought to depend on a hierarchy of abstraction in the lateral PFC along the rostral-caudal axis (Badre and D'Esposito, 2007; Badre and D'Esposito, 2009). At the apex of this hierarchy is the rostrolateral PFC (rlPFC), a region often implicated in tasks that require people to generalize across abstract, symbolic representations. For example, relational reasoning tasks such as Raven’s progressive matrices and rule-based tasks involving abstract relations are thought to depend on left rlPFC (Christoff et al., 2001; Bunge et al., 2005, Bunge et al., 2009; Davis et al., 2017). In addition to its role in generalizing abstract, relational rules, we have recently found left rlPFC to be involved in rule evaluation and novel generalization processes for simpler feature-based rules in categorization tasks (Paniukov and Davis, 2018). In the present study, dissimilarity-based generalization to novel feature pairings may depend on rule evaluation processes in the rlPFC more so than simple similarity-based processing, if studies anticipating that dissimilarity-based processes depend more upon higher-level symbolic rules are correct (Juslin et al., 2001; Winman et al., 2005). Alternatively, pure similarity-based accounts suggest that generalization patterns in an IBRE task do not depend on the existence of a separate, higher-level mechanism (Medin and Edelson, 1988; Kruschke, 1996, Kruschke, 2001), and would thus expect a single neurobiological network associated with similarity-based processing to be engaged for choice across trials.

Here we test the dissGCM by incorporating its predictions into an analysis of fMRI data collected from participants completing a standard IBRE task (Medin and Edelson, 1988; Kruschke, 1996). We first examine whether multivoxel activation patterns elicited during conflicting trials in the IBRE task are consistent with participants activating information associated with the rare category, as predicted by pure similarity-based accounts, or activating information associated with (dissimilarity to) the common category, as predicted by the dissGCM. To this end, we use representational similarity analysis (RSA; Kriegeskorte et al., 2008) to decode which features of the stimuli are most strongly activated while participants are categorizing the conflicting items. This analysis is based on recent work in the broader memory literature establishing that it is possible to decode whether participants are retrieving particular object categories from memory based on their activation patterns in ventral temporal cortex (Rissman and Wagner, 2012; Haxby et al., 2014).

To facilitate the multivoxel analysis, here we use the real world visual categories faces, scenes, and objects as stimulus features. These visual categories have a well-defined representational topography across the cortex (Haxby et al., 2001; Grill-Spector and Weiner, 2014), allowing us to predict whether participants are differentially activating particular stimulus features (faces, scenes, or objects) by computing similarities between activation patterns elicited for the key IBRE trials and feature-specific patterns from an independent localizer scan. By crossing the visual stimulus features with our category structure (Figure 1), we create situations where a rare category is associated with one feature type (e.g. a scene) and a common category is associated with another feature type (e.g. an object). The extent to which each type of information is active can then be compared to determine whether participants are representing stimulus features associated with the common or rare category on a trial, and thus answer whether their BOLD activation patterns are more consistent with pure similarity or dissGCM’s combined dissimilarity and similarity processes. In this context, we anticipate that the multivoxel pattern analysis will index an interactive process between feature-based attention and memory retrieval: the dissGCM, pure similarity-based GCM, and previous dissimilarity-based inference models all predict that categorization decisions are driven by an attention-weighted memory process whereby a stimulus is compared with the contents of memory (Nosofsky, 1986; Juslin et al., 2001; Stewart and Morin, 2007). This prediction suggests that during categorization, the multivoxel patterns activated for a particular stimulus will reflect both direct perceptual processing and retrieval of information from memory. Because the dissGCM predicts greater contributions from the common, unchosen category during this retrieval process, we expect multivoxel patterns during ambiguous trials to reveal greater activation of information associated with the common, unchosen category.

Figure 1

Download asset Open asset

In addition to our multivoxel analysis, we also test whether using dissimilarity-based evidence against unchosen categories may tap distinct brain regions, such as the rlPFC, beyond those involved with similarity-based computations. To this end, we take trial-by-trial predictions for how much similarity- and dissimilarity-based evidence contribute to the winning category and use these predictions as regressors in fMRI analysis. We anticipated that the MTL and vmPFC would be positively associated with similarity-based evidence, whereas dlPFC would be negatively associated with similarity-based evidence for the winning category. Contrastingly, we expected rlPFC to track estimates of dissimilarity-based evidence against alternative options.

The present study employed model-based fMRI to test how similarity and dissimilarity contribute to the inverse base rate effect (IBRE) and how these types of evidence relate to neural mechanisms that support category learning. The dominant theory behind the IBRE suggests that it arises from attentional processes that make ambiguous items containing features of rare and common categories seem more similar to members of the rare category. Here we find support for the hypothesis that dissimilarity-based evidence also contributes to the IBRE: people may categorize the ambiguous stimuli as members of the rare category not only because of their similarity to the rare category, but also because of their dissimilarity to members of the common category.

The dissGCM, an extension of the GCM that allows for the use of dissimilarity-based evidence in categorization behavior, predicted two novel observations in the neuroimaging data. First, as predicted by the dissGCM’s relative contribution of similarity- and dissimilarity-based evidence during the ambiguous trials, multivoxel analysis suggested stronger activation of patterns associated with features of the common category when participants classified ambiguous stimuli as rare. Second, model-based univariate analysis revealed that measures of similarity- and dissimilarity-based evidence had unique neural topographies. Similarity-based evidence for the winning category was positively correlated with regions of the hippocampus, vmPFC, and primary motor cortex. In contrast, dlPFC, dorsomedial PFC, and posterior parietal cortex were negatively correlated with similarity-based contributions. Dissimilarity-based evidence against non-winning categories was positively correlated with the left rlPFC.

The present results raise several important questions about the cognitive and neural mechanisms underlying people’s use of base rate information. Previous theories arguing for dissimilarity-like processes as explanations of IBRE have argued that they arise from mechanisms rooted in higher-level propositional logic that fundamentally differ from the similarity-based mechanisms posited by dominant theories (Juslin et al., 2001). As illustrated by the dissGCM, such dissimilarity-based processes can be viewed as simple extensions of similarity-based processing and need not depend on the existence of a functionally separate categorization system. At the same time, our neuroimaging results suggest that dissimilarity, but not similarity-based evidence may arise from processing in rlPFC regions that are known to be involved with higher-level reasoning and problem solving (Christoff et al., 2001; Bunge et al., 2005, Bunge et al., 2009). One possibility for reconciling these theories is that the dissimilarity-based evidence involves more abstract or symbolic feature processing than pure similarity processes, and this additional processing taps rlPFC regions. This is consistent with our recent model-based fMRI results, which demonstrate that rlPFC tracks measures of relational encoding in category learning, but otherwise this type of category learning may rely on the same basic similarity-based mechanisms as simpler feature-based learning (Davis et al., 2017).

By establishing that the rlPFC is engaged when participants incorporate dissimilarity-based evidence into categorization decisions, our research adds to a growing literature aiming to pinpoint a domain-general computational role for this region. A common thread among tasks shown to engage the rlPFC is that they tend to involve combining across disparate representations to form the basis for a decision – whether those representations comprise confidence estimates and subjective value (De Martino et al., 2013), visual features and their relations (Bunge, 2004; Bunge et al., 2009; Davis et al., 2017), or expected rewards and their relative uncertainties (Boorman et al., 2009; Badre et al., 2012). Likewise, in the case of the current study, the evidence that an ambiguous stimulus is similar to a given category must be combined with the evidence that the stimulus is dissimilar to the other possible categories. Although the dissGCM instantiates dissimilarity as a simple transformation of similarity, the involvement of rlPFC when participants place more reliance on dissimilarity-based evidence may be attributable to increasing demands for integrating evidence across several abstract representations. A decision made on pure similarity-based evidence would require no such integration. This hypothesis accords with recent findings implicating the rlPFC in evaluative processes for categorization tasks that require candidate rules to be weighed over the course of several trials, relative to matching tasks where a rule can be known with certainty following a single correct trial (Paniukov and Davis, 2018).

One question that has arisen repeatedly in the literature on the IBRE is whether it reflects an inherent irrationality in decision making. When viewed through the lens of basic similarity-based attentional processes (e.g. Medin and Edelson, 1988; Kruschke, 1996, Kruschke, 2001), the IBRE appears to arise from very simple learning mechanisms that are not particularly tied to higher-level rationality, and rare choices seem to indicate a lack of knowledge of the base rates. Indeed, in a separate model fit, we attempted to fit the standard similarity-based GCM to the key pattern on the ambiguous trials. However, the standard GCM was only able to predict a greater proportion of rare choices if accurate knowledge of the exemplar base rates was eliminated (all values of t_j = 1 or fit as free parameters). In contrast, accurate knowledge of the category base rates directly contributes to the greater dissimilarity-based evidence against the common category. Thus from the dissGCM perspective, participants are perfectly knowledgeable about the base rates in the present task, but they use this knowledge in a way not anticipated by pure similarity-based models. However, whether or not this use of dissimilarity-based evidence constitutes irrationality is a deeper question that cannot be answered based purely on the present results.

How or whether the use of dissimilarity is encouraged by the standard IBRE design, compared with other types of categorization problems, is an open question. The dissGCM was originally developed to explain sequential effects in categorization (Stewart and Morin, 2007), and its success in this domain suggests that dissimilarity processes, such as those revealed here, may be present in many categorization tasks that are more familiar in the neuroimaging literature. However, how much a task encourages dissimilarity-based processing may vary considerably and depend on a number of factors. For example, purely attentional accounts posit that strong initial learning of the common category leads people to learn the rare category by the features that distinguish it from the common (Markman, 1989; Kruschke, 1996). Learning-order effects do appear to play a role in the IBRE: previous studies have shown that using blocked or unequally distributed category frequencies during training leads participants to favor later-learned categories on ambiguous test probes, even when overall base rates are held constant (Medin and Bettger, 1991; Kruschke, 1996).

Early studies on the role of order and blocking on IBRE are similar to current work on blocked versus interleaved learning in the broader categorization literature (Birnbaum et al., 2013; Carvalho and Goldstone, 2014, Carvalho and Goldstone, 2015; Goldwater et al., 2018). In blocked learning, categories are learned by viewing a number of items from the same category before switching to other categories. For example, in the present case, if we had used blocked learning, participants may see an entire block of Disease 1 examples, and then blocks of Disease 2, 3, and 4, but the examples of the Diseases would not be intermixed. In interleaved learning, the standard for category learning, items from all categories are presented in a random order such that the examples of the different categories are intermixed. Blocked learning tends to lead participants to focus more on stimulus features that are shared with members of the same category, whereas interleaved learning tends to lead participants to focus more on features that differentiate categories. Interestingly, while not an interleaved versus blocked manipulation per se, frequency manipulations such as those used in the present study have an effect of creating more blocking within common categories – common categories are more likely to follow examples of the same category, and interleaving within rare categories. Although discovered long after the initial IBRE studies, blocked versus interleaved learning theories may offer a concurring explanation of the IBRE that does not depend on differences in the rate at which common and rare categories are learned. However, formal computational models of blocked versus interleaved learning have thus far focused on how these scenarios produce differences in selective attention to stimulus features that are characteristic (blocked) or diagnostic (interleaved) of a category, and are pure similarity-based models such as the original GCM (Carvalho and Goldstone, 2017). Contrastingly, our MVPA and univariate fMRI results show that pure similarity-based processing cannot fully explain the IBRE, and thus strongly suggest that dissimilarity processes contribute to the IBRE.

To investigate how learning manipulations, such as blocked versus interleaved, or individual differences in learning influence the mechanisms we propose here, it will be critical for future research to build full learning models of the dissGCM. The dissGCM, like the standard GCM, is a model of asymptotic categorization performance and generalization, and thus is not well-equipped to account for learning dynamics or individual differences. For these reasons, our individual difference analysis focused on using MVPA estimates from individual stimuli rather than formal model-based analysis. Nonetheless, these analyses reveal important individual differences that are consistent with dissimilarity-based theories more broadly. Dissimilarity-based theories posit that one of the reasons IBRE arises is because the common category becomes more thoroughly established in memory during learning, which leads participants to retrieve this information more readily at test. From this perspective, participants who learn the common category more strongly should consequently exhibit base rate neglect more frequently. Consistent with these predictions of dissimilarity-based theories, pattern similarity analyses revealed that participants who more strongly activated information associated with common categories during learning engaged in base rate neglect more often at test. While these results suggest that individual differences in learning contribute to IBRE, they nonetheless point to a critical need to develop a full learning-based version of the dissGCM that can be applied at the individual level to capture these differences. For example, one possibility is that participants’ weighting of similarity and dissimilarity (the s parameter) changes over learning based on the participants’ learning rates and factors related to blocking versus interleaved presentation (Carvalho and Goldstone, 2017). However, such a model would require extensive additional data to validate, and thus is beyond the scope of the present study.

The IBRE exemplifies a case in cognitive neuroscience where independent models that predict essentially the same behavioral patterns make very different assumptions about the cognitive processes, and accordingly, brain states, involved in producing the behavior. Our findings from the test phase represent a critical step forward in an emerging area of research using multivariate fMRI to reveal that qualitatively distinct brain states may reflect the use of multiple response strategies in the face of identical stimuli (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. Moreover, rare category selections were accompanied by slower RTs relative to common selections. These results are consistent with a higher-level, dissimilarity-based process where activating information associated with common exemplars provides contrastive evidence against the well-established common category. Alternatively, it is possible that participants are more likely to respond according to the base rates when the ambiguous stimuli elicit a strong similarity-based match: given our RT results along with the correlation between similarity-based evidence and motor cortex engagement, in these cases subjects may revert to habitual response patterns from the learning phase and simply choose the more well-established (common) category. However, understanding the precise cognitive mechanisms that contribute to these response-dependent activation patterns remains a direction for future research.

Interestingly, while our findings argue against the prediction from similarity-based models that the IBRE arises because rare features become more similar to their associated category, the observed attention weight parameters w_k from the model fits are consistent with a key part of similarity theory – that there is greater selective attention allocated to the rare feature dimension. Indeed, the rare feature dimensions outweighed the common features for both sets of categories in our data. However, these larger attention weights did not seem to drive greater pattern similarity to the rare feature dimension in our multivoxel results. 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). Rather than adjudicating between whether the multivoxel patterns in the current study are more likely to indicate attention or memory, a possibility that accords with both potential explanations is that these pattern similarity indices reflect information that is actively represented in working memory, either by way of visual cueing or reinstated long-term memories (Lewis-Peacock and Postle, 2008). In cases in which multiple or competing stimulus representations are present in WM, as may be expected for the ambiguous IBRE trials, multivoxel patterns should be most similar to whichever representation is consciously attended (Lewis-Peacock et al., 2012). However, given the design of the current study we are unable to rule out the possibility that implicitly activated or post-decisional feature representations contribute to our pattern similarity results. Future studies may wish to combine multivoxel pattern analysis with eye-tracking (e.g. Leong et al., 2017) to better understand the unique contributions that attention and memory make to the present results.

In conclusion, using model-based fMRI analysis, we found evidence that extreme cases of base rate neglect such as the IBRE may arise from a combination of similarity- and dissimilarity-based processes. Accordingly, measures of neural activation suggest that people may be more strongly relying on evidence about how dissimilar an item is to common categories when faced with ambiguous stimuli. Furthermore, dissimilarity processes have a unique cortical topography that includes the rostrolateral PFC, a region believed to be involved with more symbolic feature processing.

Twenty-four healthy right-handed volunteers (age range 18–58; 13 women) participated in the study for $35. All protocols were approved by the Texas Tech University IRB. Two participants were excluded, one for falling asleep and the other for registration failures in the first five scanning runs.

Source data and scripts used to create all figures and tables (e.g. R code, PyMVPA scripts, statistical maps for the model-based fMRI analysis) are posted to a publicly available online repository (Open Science Framework: https://osf.io/atbz7/). Raw fMRI data for the study organized according to Brain Imaging Data Structure (BIDS) guidelines are available at https://openneuro.org/datasets/ds001302.

1. 1. Sean R O'Bryan

   (2018) Model-based fMRI reveals dissimilarity processes underlying base rate neglect

   Publicly available at the Open Science Framework.

   https://osf.io/atbz7/
2. 1. Sean R O'Bryan
   2. Darrell A Worthy
   3. Evan J Livesey
   4. Tyler Davis

   (2018) Inverse Base Rate

   Publicly available at the Open Neuro website (accession no. ds001302).

   https://openneuro.org/datasets/ds001302

1. 1. Badre D
   2. D'Esposito M

   (2007) Functional magnetic resonance imaging evidence for a hierarchical organization of the prefrontal cortex

   Journal of Cognitive Neuroscience 19:2082–2099.

   https://doi.org/10.1162/jocn.2007.19.12.2082

   • PubMed
   • Google Scholar
2. 1. Badre D
   2. D'Esposito M

   (2009) Is the rostro-caudal axis of the frontal lobe hierarchical?

   Nature Reviews Neuroscience 10:659–669.

   https://doi.org/10.1038/nrn2667

   • PubMed
   • Google Scholar
3. 1. Badre D
   2. Doll BB
   3. Long NM
   4. Frank MJ

   (2012) Rostrolateral prefrontal cortex and individual differences in uncertainty-driven exploration

   Neuron 73:595–607.

   https://doi.org/10.1016/j.neuron.2011.12.025

   • PubMed
   • Google Scholar
4. 1. Birnbaum MS
   2. Kornell N
   3. Bjork EL
   4. Bjork RA

   (2013) Why interleaving enhances inductive learning: the roles of discrimination and retrieval

   Memory & Cognition 41:392–402.

   https://doi.org/10.3758/s13421-012-0272-7

   • PubMed
   • Google Scholar
5. 1. Boettiger CA
   2. D'Esposito M

   (2005) Frontal networks for learning and executing arbitrary stimulus-response associations

   Journal of Neuroscience 25:2723–2732.

   https://doi.org/10.1523/JNEUROSCI.3697-04.2005

   • PubMed
   • Google Scholar
6. 1. Boorman ED
   2. Behrens TE
   3. Woolrich MW
   4. Rushworth MF

   (2009) How green is the grass on the other side? Frontopolar cortex and the evidence in favor of alternative courses of action

   Neuron 62:733–743.

   https://doi.org/10.1016/j.neuron.2009.05.014

   • PubMed
   • Google Scholar
7. 1. Bravata DM

   (2000)

   Making medical decisions under uncertainty

   Seminars in Medical Practice 3:6–13.

   • Google Scholar
8. 1. Bunge SA
   2. Helskog EH
   3. Wendelken C

   (2009) Left, but not right, rostrolateral prefrontal cortex meets a stringent test of the relational integration hypothesis

   NeuroImage 46:338–342.

   https://doi.org/10.1016/j.neuroimage.2009.01.064

   • PubMed
   • Google Scholar
9. 1. Bunge SA
   2. Wendelken C
   3. Badre D
   4. Wagner AD

   (2005) Analogical reasoning and prefrontal cortex: evidence for separable retrieval and integration mechanisms

   Cerebral Cortex 15:239–249.

   https://doi.org/10.1093/cercor/bhh126

   • PubMed
   • Google Scholar
10. 1. Bunge SA

    (2004) How we use rules to select actions: a review of evidence from cognitive neuroscience

    Cognitive, Affective, & Behavioral Neuroscience 4:564–579.

    https://doi.org/10.3758/CABN.4.4.564

    • PubMed
    • Google Scholar
11. 1. Carvalho PF
    2. Goldstone RL

    (2014) Putting category learning in order: category structure and temporal arrangement affect the benefit of interleaved over blocked study

    Memory & Cognition 42:481–495.

    https://doi.org/10.3758/s13421-013-0371-0

    • PubMed
    • Google Scholar
12. 1. Carvalho PF
    2. Goldstone RL

    (2015) The benefits of interleaved and blocked study: different tasks benefit from different schedules of study

    Psychonomic Bulletin & Review 22:281–288.

    https://doi.org/10.3758/s13423-014-0676-4

    • PubMed
    • Google Scholar
13. 1. Carvalho PF
    2. Goldstone RL

    (2017) The sequence of study changes what information is attended to, encoded, and remembered during category learning

    Journal of Experimental Psychology: Learning, Memory, and Cognition 43:1699–1719.

    https://doi.org/10.1037/xlm0000406

    • Google Scholar
14. 1. Casscells W
    2. Schoenberger A
    3. Graboys TB

    (1978) Interpretation by physicians of clinical laboratory results

    New England Journal of Medicine 299:999–1001.

    https://doi.org/10.1056/NEJM197811022991808

    • PubMed
    • Google Scholar
15. 1. Christoff K
    2. Prabhakaran V
    3. Dorfman J
    4. Zhao Z
    5. Kroger JK
    6. Holyoak KJ
    7. Gabrieli JD

    (2001) Rostrolateral prefrontal cortex involvement in relational integration during reasoning

    NeuroImage 14:1136–1149.

    https://doi.org/10.1006/nimg.2001.0922

    • PubMed
    • Google Scholar
16. 1. Davis T
    2. Goldwater M
    3. Giron J

    (2017) From concrete examples to abstract relations: the rostrolateral prefrontal cortex integrates novel examples into relational categories

    Cerebral Cortex 27:2652–2670.

    https://doi.org/10.1093/cercor/bhw099

    • PubMed
    • Google Scholar
17. 1. Davis T
    2. Love BC
    3. Preston AR

    (2012a) Learning the exception to the rule: model-based FMRI reveals specialized representations for surprising category members

    Cerebral Cortex 22:260–273.

    https://doi.org/10.1093/cercor/bhr036

    • PubMed
    • Google Scholar
18. 1. Davis T
    2. Love BC
    3. Preston AR

    (2012b) Striatal and hippocampal entropy and recognition signals in category learning: simultaneous processes revealed by model-based fMRI

    Journal of Experimental Psychology: Learning, Memory, and Cognition 38:821–839.

    https://doi.org/10.1037/a0027865

    • Google Scholar
19. 1. Davis T
    2. Poldrack RA

    (2014) Quantifying the internal structure of categories using a neural typicality measure

    Cerebral Cortex 24:1720–1737.

    https://doi.org/10.1093/cercor/bht014

    • PubMed
    • Google Scholar
20. 1. De Martino B
    2. Fleming SM
    3. Garrett N
    4. Dolan RJ

    (2013) Confidence in value-based choice

    Nature Neuroscience 16:105–110.

    https://doi.org/10.1038/nn.3279

    • PubMed
    • Google Scholar
21. 1. DeGutis J
    2. D'Esposito M

    (2007) Distinct mechanisms in visual category learning

    Cognitive, Affective, & Behavioral Neuroscience 7:251–259.

    https://doi.org/10.3758/CABN.7.3.251

    • PubMed
    • Google Scholar
22. 1. Filoteo JV
    2. Maddox WT
    3. Simmons AN
    4. Ing AD
    5. Cagigas XE
    6. Matthews S
    7. Paulus MP

    (2005) Cortical and subcortical brain regions involved in rule-based category learning

    NeuroReport 16:111–115.

    https://doi.org/10.1097/00001756-200502080-00007

    • PubMed
    • Google Scholar
23. 1. Goldwater MB
    2. Don HJ
    3. Krusche MJF
    4. Livesey EJ

    (2018) Relational discovery in category learning

    Journal of Experimental Psychology: General 147:1–35.

    https://doi.org/10.1037/xge0000387

    • PubMed
    • Google Scholar
24. 1. Grill-Spector K
    2. Weiner KS

    (2014) The functional architecture of the ventral temporal cortex and its role in categorization

    Nature Reviews Neuroscience 15:536–548.

    https://doi.org/10.1038/nrn3747

    • PubMed
    • Google Scholar
25. 1. Hanke M
    2. Halchenko YO
    3. Sederberg PB
    4. Hanson SJ
    5. Haxby JV
    6. Pollmann S

    (2009) PyMVPA: A python toolbox for multivariate pattern analysis of fMRI data

    Neuroinformatics 7:37–53.

    https://doi.org/10.1007/s12021-008-9041-y

    • PubMed
    • Google Scholar
26. 1. Haxby JV
    2. Connolly AC
    3. Guntupalli JS

    (2014) Decoding neural representational spaces using multivariate pattern analysis

    Annual Review of Neuroscience 37:435–456.

    https://doi.org/10.1146/annurev-neuro-062012-170325

    • PubMed
    • Google Scholar
27. 1. Haxby JV
    2. Gobbini MI
    3. Furey ML
    4. Ishai A
    5. Schouten JL
    6. Pietrini P

    (2001) Distributed and overlapping representations of faces and objects in ventral temporal cortex

    Science 293:2425–2430.

    https://doi.org/10.1126/science.1063736

    • PubMed
    • Google Scholar
28. 1. Ishai A
    2. Ungerleider LG
    3. Martin A
    4. Schouten JL
    5. Haxby JV

    (1999) Distributed representation of objects in the human ventral visual pathway

    PNAS 96:9379–9384.

    https://doi.org/10.1073/pnas.96.16.9379

    • PubMed
    • Google Scholar
29. 1. Juslin P
    2. Wennerholm P
    3. Winman A

    (2001) High-level reasoning and base-rate use: do we need cue-competition to explain the inverse base-rate effect?

    Journal of Experimental Psychology: Learning, Memory, and Cognition 27:849–871.

    https://doi.org/10.1037/0278-7393.27.3.849

    • PubMed
    • Google Scholar
30. 1. Kriegeskorte N
    2. Mur M
    3. Bandettini P

    (2008) Representational similarity analysis - connecting the branches of systems neuroscience

    Frontiers in Systems Neuroscience 2:4.

    https://doi.org/10.3389/neuro.06.004.2008

    • PubMed
    • Google Scholar
31. 1. Kruschke JK

    (1996) Base rates in category learning

    Journal of Experimental Psychology: Learning, Memory, and Cognition 22:3–26.

    https://doi.org/10.1037/0278-7393.22.1.3

    • Google Scholar
32. 1. Kruschke JK

    (2001) Toward a unified model of attention in associative learning

    Journal of Mathematical Psychology 45:812–863.

    https://doi.org/10.1006/jmps.2000.1354

    • Google Scholar
33. 1. Leong YC
    2. Radulescu A
    3. Daniel R
    4. DeWoskin V
    5. Niv Y

    (2017) Dynamic interaction between reinforcement learning and attention in multidimensional environments

    Neuron 93:451–463.

    https://doi.org/10.1016/j.neuron.2016.12.040

    • PubMed
    • Google Scholar
34. 1. Lewis-Peacock JA
    2. Drysdale AT
    3. Oberauer K
    4. Postle BR

    (2012) Neural evidence for a distinction between short-term memory and the focus of attention

    Journal of Cognitive Neuroscience 24:61–79.

    https://doi.org/10.1162/jocn_a_00140

    • PubMed
    • Google Scholar
35. 1. Lewis-Peacock JA
    2. Postle BR

    (2008) Temporary activation of long-term memory supports working memory

    Journal of Neuroscience 28:8765–8771.

    https://doi.org/10.1523/JNEUROSCI.1953-08.2008

    • PubMed
    • Google Scholar
36. 1. Liu Z
    2. Braunlich K
    3. Wehe HS
    4. Seger CA

    (2015) Neural networks supporting switching, hypothesis testing, and rule application

    Neuropsychologia 77:19–34.

    https://doi.org/10.1016/j.neuropsychologia.2015.07.019

    • PubMed
    • Google Scholar
37. 1. Mack ML
    2. Love BC
    3. Preston AR

    (2016) Dynamic updating of hippocampal object representations reflects new conceptual knowledge

    PNAS 113:13203–13208.

    https://doi.org/10.1073/pnas.1614048113

    • PubMed
    • Google Scholar
38. 1. Mack ML
    2. Preston AR
    3. Love BC

    (2013) Decoding the brain's algorithm for categorization from its neural implementation

    Current Biology 23:2023–2027.

    https://doi.org/10.1016/j.cub.2013.08.035

    • PubMed
    • Google Scholar
39. 1. Markman AB

    (1989) LMS rules and the inverse base-rate effect: comment on Gluck and Bower (1988)

    Journal of Experimental Psychology: General 118:417–421.

    https://doi.org/10.1037/0096-3445.118.4.417

    • Google Scholar
40. 1. Medin DL
    2. Bettger JG

    (1991) Sensitivity to changes in Base-Rate information

    The American Journal of Psychology 104:311–332.

    https://doi.org/10.2307/1423242

    • Google Scholar
41. 1. Medin DL
    2. Edelson SM

    (1988) Problem structure and the use of base-rate information from experience

    Journal of Experimental Psychology: General 117:68–85.

    https://doi.org/10.1037/0096-3445.117.1.68

    • PubMed
    • Google Scholar
42. 1. Mumford JA
    2. Turner BO
    3. Ashby FG
    4. Poldrack RA

    (2012) Deconvolving BOLD activation in event-related designs for multivoxel pattern classification analyses

    NeuroImage 59:2636–2643.

    https://doi.org/10.1016/j.neuroimage.2011.08.076

    • PubMed
    • Google Scholar
43. 1. Nasser HM
    2. Calu DJ
    3. Schoenbaum G
    4. Sharpe MJ

    (2017) The dopamine prediction error: contributions to associative models of reward learning

    Frontiers in Psychology 8:244.

    https://doi.org/10.3389/fpsyg.2017.00244

    • PubMed
    • Google Scholar
44. 1. Nosofsky RM

    (1986) Attention, similarity, and the identification-categorization relationship

    Journal of Experimental Psychology: General 115:39–57.

    https://doi.org/10.1037/0096-3445.115.1.39

    • PubMed
    • Google Scholar
45. 1. O'Bryan SR
    2. Walden E
    3. Serra MJ
    4. Davis T

    (2018) Rule activation and ventromedial prefrontal engagement support accurate stopping in self-paced learning

    NeuroImage 172:415–426.

    https://doi.org/10.1016/j.neuroimage.2018.01.084

    • PubMed
    • Google Scholar
46. 1. Paniukov D
    2. Davis T

    (2018) The evaluative role of rostrolateral prefrontal cortex in rule-based category learning

    NeuroImage 166:19–31.

    https://doi.org/10.1016/j.neuroimage.2017.10.057

    • PubMed
    • Google Scholar
47. 1. Rissman J
    2. Gazzaley A
    3. D'Esposito M

    (2004) Measuring functional connectivity during distinct stages of a cognitive task

    NeuroImage 23:752–763.

    https://doi.org/10.1016/j.neuroimage.2004.06.035

    • PubMed
    • Google Scholar
48. 1. Rissman J
    2. Wagner AD

    (2012) Distributed representations in memory: insights from functional brain imaging

    Annual Review of Psychology 63:101–128.

    https://doi.org/10.1146/annurev-psych-120710-100344

    • PubMed
    • Google Scholar
49. 1. Seger CA
    2. Braunlich K
    3. Wehe HS
    4. Liu Z

    (2015) Generalization in category learning: the roles of representational and decisional uncertainty

    Journal of Neuroscience 35:8802–8812.

    https://doi.org/10.1523/JNEUROSCI.0654-15.2015

    • PubMed
    • Google Scholar
50. 1. Seger CA
    2. Cincotta CM

    (2006) Dynamics of frontal, striatal, and hippocampal systems during rule learning

    Cerebral Cortex 16:1546–1555.

    https://doi.org/10.1093/cercor/bhj092

    • PubMed
    • Google Scholar
51. 1. Sharpe MJ
    2. Killcross S

    (2015) The prelimbic cortex directs attention toward predictive cues during fear learning

    Learning & Memory 22:289–293.

    https://doi.org/10.1101/lm.038273.115

    • PubMed
    • Google Scholar
52. 1. Siegel JS
    2. Power JD
    3. Dubis JW
    4. Vogel AC
    5. Church JA
    6. Schlaggar BL
    7. Petersen SE

    (2014) Statistical improvements in functional magnetic resonance imaging analyses produced by censoring high-motion data points

    Human Brain Mapping 35:1981–1996.

    https://doi.org/10.1002/hbm.22307

    • PubMed
    • Google Scholar
53. 1. Soto FA
    2. Waldschmidt JG
    3. Helie S
    4. Ashby FG

    (2013) Brain activity across the development of automatic categorization: a comparison of categorization tasks using multi-voxel pattern analysis

    NeuroImage 71:284–297.

    https://doi.org/10.1016/j.neuroimage.2013.01.008

    • PubMed
    • Google Scholar
54. 1. Stewart N
    2. Morin C

    (2007) Dissimilarity is used as evidence of category membership in multidimensional perceptual categorization: a test of the similarity-dissimilarity generalized context model

    Quarterly Journal of Experimental Psychology 60:1337–1346.

    https://doi.org/10.1080/17470210701480444

    • PubMed
    • Google Scholar
55. 1. Tversky A
    2. Kahneman D

    (1974) Judgment under uncertainty: heuristics and biases

    Science 185:1124–1131.

    https://doi.org/10.1126/science.185.4157.1124

    • PubMed
    • Google Scholar
56. 1. Wagenmakers EJ

    (2007) A practical solution to the pervasive problems of p values

    Psychonomic Bulletin & Review 14:779–804.

    https://doi.org/10.3758/BF03194105

    • PubMed
    • Google Scholar
57. 1. Waldschmidt JG
    2. Ashby FG

    (2011) Cortical and striatal contributions to automaticity in information-integration categorization

    NeuroImage 56:1791–1802.

    https://doi.org/10.1016/j.neuroimage.2011.02.011

    • PubMed
    • Google Scholar
58. 1. Winman A
    2. Wennerholm P
    3. Juslin P
    4. Shanks DR

    (2005) Evidence for rule-based processes in the inverse base-rate effect

    The Quarterly Journal of Experimental Psychology Section A 58:789–815.

    https://doi.org/10.1080/02724980443000331

    • Google Scholar
59. 1. Zeithamova D
    2. Dominick AL
    3. Preston AR

    (2012) Hippocampal and ventral medial prefrontal activation during retrieval-mediated learning supports novel inference

    Neuron 75:168–179.

    https://doi.org/10.1016/j.neuron.2012.05.010

    • PubMed
    • Google Scholar

Author details

1. Sean R O'Bryan

   Department of Psychological Sciences, Texas Tech University, Lubbock, United States

   Contribution

   Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing

   For correspondence

   sean.r.obryan@ttu.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-0562-8211

2. Darrell A Worthy

   Department of Psychology, Texas A&M University, College Station, United States

   Contribution

   Conceptualization, Investigation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

3. Evan J Livesey

   School of Psychology, University of Sydney, Sydney, Australia

   Contribution

   Conceptualization, Investigation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

4. Tyler Davis

   Department of Psychological Sciences, Texas Tech University, Lubbock, United States

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing

   Competing interests

   No competing interests declared

• 1,088

  views

• 138

  downloads

• 10

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.