This thesis examined the role of procedural learning in human probabilistic category learning (PCL). It was proposed that there was a lack of clear behavioural evidence for learning without awareness in PCL. Eleven experiments are reported that investigated the characteristics of learning in a prototypical probabilistic category learning task (the weather prediction task). The results were interpreted as contradicting the popular interpretation of weather prediction task learning as procedurally based. Rather, it was shown that behavioural data was consistent with declarative learning. This learning was not dissociable with measures of cue knowledge. Strategy analysis converged with the behavioural data, suggesting the dominance of declarative learning in this task. It was proposed that a single system account (e.g., Lagnado et al., 2006; Newell et al., 2007) which does not posit a role for procedural learning was the most appropriate way to understand learning in the weather prediction task.

This thesis examined the role of procedural learning in human probabilistic category learning (PCL). It was proposed that there was a lack of clear behavioural evidence for learning without awareness in PCL. Eleven experiments are reported that investigated the characteristics of learning in a prototypical probabilistic category learning task (the weather prediction task). The results were interpreted as contradicting the popular interpretation of weather prediction task learning as procedurally based. Rather, it was shown that behavioural data was consistent with declarative learning. This learning was not dissociable with measures of cue knowledge. Strategy analysis converged with the behavioural data, suggesting the dominance of declarative learning in this task. It was proposed that a single system account (e.g., Lagnado et al., 2006; Newell et al., 2007) which does not posit a role for procedural learning was the most appropriate way to understand learning in the weather prediction task.

Trial-level evidence for feedback sensitivity in fixations during category learning has been previously described as weak. In this dissertation, steps were taken to overcome some methodological issues potentially obscuring the evidence for such sensitivity. Jointly, the three experiments reported here suggest that sensitivity to error in visual attention reflects cue competition, as opposed to error-driven learning of a selective visual profile. These outcomes are in agreement with previous research in human vision, which holds that fixations reflect the agent’s task representation. A case is made for the top-down control of visual attention during category learning, manifested as effects of prior knowledge, long-standing expectations, decisional uncertainty, and vacillations between alternative sources of conflicting evidence. A suggestion is made that the time-based measures of visual attention may align with the continuous ratings of the perceived category membership (reflecting learner confidence).

The field of category learning is replete with theories that detail how similarity and comparison based processes are used to learn categories, but these theories are limited to cases in which item and category representations consist of feature vectors. This precludes these methods from learning relational categories, where membership is determined by the structured relations binding the features of a stimulus together. Fortuitously,  researchers within the analogy literature have developed theories of comparison that account for this structure.  This thesis bridges the two approaches, describing a theory of category learning that utilizes the representational frameworks provided by the analogy literature to learn categories that may only be described through the appreciation of the structured relations within their members. This theory is formalized in a model, Building Relations through Instance Driven Gradient Error Shifting (BRIDGES), that shows how relational categories can be learned through attention-driven analogies between concrete exemplars.  This approach is demonstrated through several simulations that compare similarity-based learning and alternatives, such as rule-based abstractions and re-representation.  We then present a series of experiments that explore the reciprocal impact of relational comparison on category structure and category structure on relational comparison.  This work provides a theoretical framework and formal model suggesting that feature-based and relation-based categories are a continuum that are learned through selective attention and similarity-based comparison. / text

Categorization is a fundamental process that underlies much of cognition. People form categories that allow them to generalize to and make inferences about novel objects and events. Current accounts of category learning suggest that there are two systems for learning categories, an explicit rule-based system that depends on frontal-striatal loops and working memory, and a procedural system that learns implicitly and depends on the tail of the caudate nucleus and occipital regions. In the present thesis, I propose that an additional declarative category learning system exists that is recruited to learn categories that are associated with multiple conjunctive and explicit, but not strictly rule-based, representations. The basis of the declarative category learning system is then tested in several behavioral and physiological recording experiments. The first issue that is examined in relation to the declarative category learning system is how subjects’ ability to encode stimuli affects their ability to form new flexible conjunctive representations. I provide evidence consistent with the idea that there are two ways to encode stimuli in category learning, either as a conjunction of individual parts or as holistic images. Forming part-based representations is found to be especially critical for forming new conjunctive representations for exceptions in brief single session experiments. A second question is how emotional processes interact with the declarative category learning system. Numerous lines of evidence suggest that emotional processes strongly affect learning and behavior. In a study using skin conductance, I find that anticipatory emotions (i.e., emotions present before a behavioral response) show a pattern consistent with orienting attention to behaviorally significant or potentially novel events. A final fMRI project ties together hypotheses about anticipatory emotions and encoding to their neural basis and provides a test of the predicted mapping of the declarative category learning system to the brain. By relating quantitative predictions from SUSTAIN, a model that shares relationships to the medial temporal lobes (MTL) and declarative category learning system, to fMRI data, I find clusters in an MTL-midbrain-PFC network that show patterns of activation consistent with recognizing exception items and updating these representations in response to error or surprise. / text

The objective of this thesis is to examine the validity of functional near-infrared spectroscopy (fNIR) to examine brain regions involved in rule based (RB) and information integration (II) category learning. We predicted similar patterns of activation found by past studies that used fMRI scans. Our goal was to test if fNIR would be able to detect changes in blood oxygenation levels of participants who learned to categorize (learners) vs. those that did not (non learners). The stimulus set comprised of lines that differed in length and orientation. Participants had to learn to categorize by trial and error based on the feedback provided. Behavioral and neuroimaging data was recorded for both RB and II conditions. Results showed an upward trend in response accuracy over trials for participants identified as learners. Furthermore, blood oxygenation levels reported by fNIR indicated a systematic increase in oxygen consumption for learners as compared to non learners. These areas of increased prefrontal cortex activity recorded by fNIR correspond to the same areas found to be involved in categorization by fMRI. This paper reviews the background of category learning, explores various neuroimaging techniques in categorization research, and investigates the efficacy of fNIR as a relatively new neuroimaging modality by comparing it to fMRI.

No description available.

In the past decade, the COVIS model (Ashby, Alfonso-Reese, Turken, & Waldron, 1998) has emerged as the only neuropsychological theory for the existence of multiple brain systems for category learning. COVIS postulates that there are two systems, explicit and implicit, which compete against one another. These two systems reply on two discrete networks: explicit, or rule based categorization relies on executive function and working memory while implicit, or information integration categorization is mediated by dopaminergic pathways. The purpose of this pilot study was to further provide evidence for the existence of multiple systems of category learning. In all three experiments, we interrupted feedback processing using a modified Sternberg task. In Experiment 1 and 2, participants were separated into four conditions, rule based (RB) categorization with a short delay between feedback and the modified Sternberg task, RB categorization with a long delay, information integration (II) categorization with a short delay, and II categorization with a long delay. Participants in the RB conditions performed worse than those in the II conditions in Experiment 1 and 2. After determining there was no significant difference between the short and long delay manipulations, only the short delay was used for Experiment 3. Consistent with Experiment 1 and 2, participants in the RB condition performed worse than those in the II condition. Functional near-infrared spectroscopy (fNIRS) technology was also used in Experiment 3 to determine the difference in prefrontal activation between RB and II conditions. Although statistically not significant, across blocks, the difference in prefrontal activation increased.

No description available.

No description available.