Schrödinger’s Category: Active Learning in the Face of Label Ambiguity

AbstractResearch on active category learning has routinely shown that people prefer to sample regions of the space with high uncertainty (near category boundaries). Prevailing accounts suggest this strategy facilitates an understanding of the subtle distinctions between categories. However, prior work has focused on situations where category boundaries are rigid. Yet, boundaries between natural categories are often fuzzy. We ask: Do learners pursue uncertainty sampling when labels at the boundary are themselves uncertain? We introduce a fuzzy buffer around a target category that returns conflicting labels from two ‘teachers,’ and evaluate how sampling and representation are affected. We find that, relative to the rigid boundary case, learners avoid uncertainty, opting to sample densely from highly certain regions of the target category rather than its boundary. Subsequent generalization tests reveal that the sampling strategies encouraged by the fuzzy boundary negatively affected participants' grasp of category structure, even outside the fuzzy buffer zone.

Return to previous page