Do additional features help or harm during category learning? An exploration of the curse of dimensionality in human learners

How does the number of features impact category learning? One view suggests that additional features creates a “curse of dimensionality” where having more features causes the size of the search space to grow so quickly that discovering good classification rules becomes increasingly challenging. The opposing view suggests that additional features provide a wealth of additional information which learners should be able to use to improve their classification performance. Previous research exploring this issue appears to have produced conflicting results: some find that learning improves with additional features (Hoffman & Murphy, 2006) while others find that it does not (Minda & Smith, 2001; Edgell et al., 1996). Here we investigate the possibility that category structure may explain this apparent discrepancy – that more features are useful in categories with family resemblance structure, but are not (and may even be harmful) in more rule-based categories. We find while the impact of having many features does indeed depend on category structure, the results can be explained by a single unified model: one that attends to a single feature on any given trial and uses information learned from that particular feature to make classification judgments.