4.1.1. Univariate Analysis

The categories in this experiment were designed, such that they require the subjects to discover a counting rule to differentiate between two visually similar categories and utilize a perceptual similarity strategy to identify the members of a visually distinct category. Our subjects' performance on the generalization block supports the assumption that they would recruit the proper strategies. Specifically, they would not have been able to accurately categorize novel formations into the trained categories had they exclusively relied on declarative recall of individual formations.

The superior and inferior frontal gyri were more active during the categorization of visually similar trials when compared to visually distinct trials. These regions are a part of the working memory system, where it is inferred that they are responsible for orienting attention and establishing executive control [22,51–53]. In our experiment, the subjects focused their attention toward the players on the line of scrimmage, where they were required to count each player if the formation belonged to one of the two visually similar categories. This is due, in part, to the visually similar categories having low between-category variability, which requires the engagemen<sup>t</sup> of a rule-based system. The comparison of the visually distinct category to each of the visually similar categories has much greater between-category variability and, thus, do not require the use of the cognitively taxing rule-based system [54].

Interestingly, caudate nucleus, a region that is integral to rule application, did not reach a level of significance (although this is expected given the small sample size) for the rule-based condition during training. Instead, a cluster over the hippocampus had the highest level of activation during training—a region that is well-known for its role in declarative recall [55]. It is possible that subjects utilized the rule for a short period of time during training, but relied more on the declarative recall of the few relevant players, given that subjects only needed to attend to a single feature within each stimulus to perform categorization (the number of players on the line of scrimmage). However, when encountering novel formations in the generalization block that belong to the categories acquired during training, the subjects were forced into applying the counting rule and, thus, the strong presence of caudate nucleus during generalization could reflect a more consistent reliance on rule application.

In support of our hypothesis, the robust activation of the lateral occipital cortex was present for the visually distinct category when compared to the visually similar categories. This held true throughout training and extended into the generalization block. The lateral occipital cortex has been well-established as the main region governing perceptual similarity categorization [18–21]. Perceptual similarity categorization can be carried-out with minimal working memory resources and it is optimized for instances with low within-category similarity [21]. The absence of the working memory system when subjects viewed members of the visually distinct category further supports our conclusion that this category engages the perceptual similarity system.

#### 4.1.2. Multi-Voxel Pattern Analysis

Our region-based MVPA showed that the lateral frontal and parietal regions provided the most reliable classification between the visually similar and visually distinct categories, as consistent with previous findings that rule-based categorization requires a higher degree of attentional resources. On the other hand, MVPA provides a more sensitive measure of these conditional e ffects. More specifically, MVPA provides an avenue for detecting more subtle di fferences between our conditions that lie within the activity patterns of single regions–information that is sometimes subtracted-out by traditional analyses [56]. These small activation patterns can potentially code for task-relevant information that is important to both memory systems in our experiment.

Frontoparietal regions are well-known for their importance to cognitive control; mainly selective attention to information that is relevant to the task [57,58]. However, non-human primate experiments have demonstrated that activity in the frontal and parietal regions is predictive of an array of di fferent task-relevant features, such as representations of individual stimuli, rule selection, or response selection [59–61]. Follow-up studies in humans have shown similar dissociations between stimulus sets and rules using MVPA [62,63]. These task features are essential to the rule- *and* perceptual similarity-based systems. The successful dissociation between category structures while using MVPA over frontoparietal regions in our task supports these previous findings.

#### *4.2. Experiment 2 (dEEG)*

The primary objectives of this experiment were to further determine whether multiple memory systems are recruited in a single task and evaluate the time course under which these systems are recruited. The results showed that, once the participants acquired the task, clear di fferences in the Left LIAN, MFN, and P3b components were seen between our two conditions. Overall, the amplitude of each ERP that reflected a di fference between the visually similar and visually distinct categories was largest for the similarity-based category. However, the amplitude for the right LIAN was larger for the visually similar categories, although this e ffect did not reach statistical significance. For the machine learning analysis, the classification accuracy peaked earliest in the posterior parietal region (the location of the P3b), but reliable classification could be performed while using additional electrode clusters, including the left prefrontal and medial prefrontal areas.
