Modeling the influence of variance on active search in category learning
Mon-B22-Talk III-02
Presented by: Ann-Katrin Hosch
When learning to discriminate categories in real life, people may actively search for exemplars of each category to establish a representation of the categories. We investigate the influence of category variance, i.e., by how much the exemplars vary from one another, on search length, using a novel category learning task in which participants sampled category exemplars until they felt they could categorize the objects. We manipulated the categories' variance by using a low, medium, and high level of distortion from a prototype. In two experiments, we could show that participants’ search was affected by the category’s and the counter category’s variance level.
Here, we propose a cognitive model of the search process assuming a within-category learning process and a between-category discrimination process. We assume that the within-category learning rate is determined by the variance dependent uncertainty – conversely the amount of information - participants experience while forming a category representation. Specifically, we assume that participants compare the probability distribution of a category exemplar with a uniformly distributed prototype. We formalize the discrimination process as differences in the amount of information between the two categories.
We find that the model better predicts participants' sampling behavior than a model predicting sampling based on the distortion level or a model without a discrimination process. These results indicate that search in self-regulated category learning depends on the information gained by sampling within a category and the ease with which categories can be discriminated.
Here, we propose a cognitive model of the search process assuming a within-category learning process and a between-category discrimination process. We assume that the within-category learning rate is determined by the variance dependent uncertainty – conversely the amount of information - participants experience while forming a category representation. Specifically, we assume that participants compare the probability distribution of a category exemplar with a uniformly distributed prototype. We formalize the discrimination process as differences in the amount of information between the two categories.
We find that the model better predicts participants' sampling behavior than a model predicting sampling based on the distortion level or a model without a discrimination process. These results indicate that search in self-regulated category learning depends on the information gained by sampling within a category and the ease with which categories can be discriminated.
Keywords: category learning, sampling, cognitive model, search