Investigating the Learning Mechanisms of Visual Statistical Learning

In my dissertation, I examined the mechanisms of statistical learning and the relationships that influence it. The first set of experiments investigated whether statistical learning is error-driven, using both forward and backward blocking paradigms. Although forward blocking did not show the expected blocking effect, backward blocking was observed, indicating a functional similarity between statistical learning and reinforcement learning, and supporting the idea that statistical learning may be error-driven. The second set of experiments examined whether statistical learning is more sensitive to unique predictive relationships than to conditional probabilities. The results indicated that participants learned object pairs based on unique predictive relationships, although the specific form of uniqueness remains unclear. Overall, the findings suggest that statistical learning may be error-driven, but further research is needed to determine the precise metrics of uniqueness that drive it.