UCLA

Whether it be your ability to consistently recognize the face of your friend across varying contexts, or the ability of a dermatologist to differentiate cancerous skin lesions from benign ones, many tasks and domains are reliant upon the ability to classify items into one of many possible categories. This can be challenging, especially when category members present in diverse ways or closely resemble members of other categories. Research has shown that category acquisition can be facilitated by opportunities to directly compare items together. In particular, comparisons may support rapid improvements in the discovery and pick up of information, resulting in deeper processing of category structure and enhanced category representations. The goal of this dissertation was to elucidate the most effective methods for constructing comparisons to accelerate the mechanisms of perceptual learning that underlie successful categorization. In particular, emphasis was given to exploring the value of paired comparison trials, which involved the concurrent presentation of items from two different categories for discrimination. In Experiments 1 and 2, learning trial structure and task were manipulated to test paired comparison learning against more common classification-based approaches. Results revealed that paired comparison learning was an effective way to learn the classification of a large set of categories, particularly when the domain was novel to the learner. Experiment 3 broke down paired comparisons into its separate learning components to evaluate how learning is advanced. Results revealed asymmetric learning gains in favor categories framed as the target of the trial relative to categories framed as the distractor. Experiment 4 measured the perceptual changes induced by paired comparison learning, finding evidence for within-category compression and between-category expansion. Finally, Experiment 5 evaluated adaptive learning methods to enhance comparison efficacy by testing an adaptive comparison procedure, previously used in face learning, in the domain of skin lesion classification. A partial replication of results was observed, and suggestions are given as to how the procedure may be improved. Altogether, this work has important implications for our understanding of the role of perceptual learning in high-level tasks generally, as well as for how it interacts with comparison opportunities specifically.