Adaptive categorization in unsupervised learning.

Abstract

In 3 experiments, the authors provide evidence for a distinct category-invention process in unsupervised (discovery) learning and set forth a method for observing and investigating that process. In the 1st 2 experiments, the sequencing of unlabeled training instances strongly affected participants' ability to discover patterns (categories) across those instances. In the 3rd experiment, providing diagnostic labels helped participants discover categories and improved learning even for instance sequences that were unlearnable in the earlier experiments. These results are incompatible with models that assume that people learn by incrementally tracking correlations between individual features; instead, they suggest that learners in this study used expectation failure as a trigger to invent distinct categories to represent patterns in the stimuli. The results are explained in terms of J. R. Anderson's (1990, 1991) rational model of categorization, and extensions of this analysis for real-world learning are discussed.

When more is less: Negative exposure effects in unsupervised learning

“…However, the same result can easily be accommodated within the category invention framework (Clapper & Bower, 2002). Consider a hypothetical learner exposed to two different training sequences.…”

“…To illustrate, consider an experiment reported by Clapper and Bower (2002;see also Clapper & Bower, 1994). In that experiment, participants saw a series of training instances (verbal descriptions of fictitious tree species) from two correlational patterns (potential categories), referred to here as A and B.…”

“…In particular, the correlation tracking perspective suggests that learning should be an incremental, monotonic growth process in which each repetition of a given pattern etches that pattern (or at least some of the correlations that comprise it) more deeply into memory (Clapper & Bower, 2002). Learning of a pattern can only increase, and never systematically decrease, as a function of additional exposure to that pattern (all else being equal).…”

“…Consider a hypothetical learner exposed to two different training sequences. Given a sequence similar to the first sequence above (referred to as a "contrast sequence" by Clapper & Bower, 2002), this learner would be expected to create a new category (Category A) at the beginning of the experiment (see Figure 1A) and then acquire general norms about that category by aggregating over the A instances presented during the next several trials. By the time the first B is encountered, the learner should have strong expectations regarding the consistent features of Category A, narrowing the boundaries for acceptance into that category.…”

Category labels versus feature labels: Category labels polarize inferential predictions

“…M. Markman & Hutchinson, 1984;Waxman & Booth, 2001; but see also Sloutsky, 2003;Sloutsky & Fisher, 2004). A number of cognitive studies have also demonstrated the importance of category lab bels in inductive generalization (Clapper & Bower, 2002;Waldmann & Hagmayer, 2006;Yamauchi, 2005, in press; see also Murphy & Medin, 1985;Murphy & Ross, 1994;Rosch, 1978;Ross & Murphy, 1996). However, there is no clear consensus among cognitive scientists on exactly how category labels and feature labels differ.…”