Exploring the conceptual universe.

Abstract

Humans can learn to organize many kinds of domains into categories, including real-world domains such as kinsfolk and synthetic domains such as sets of geometric figures that vary along several dimensions. Psychologists have studied many individual domains in detail, but there have been few attempts to characterize or explore the full space of possibilities. This article provides a formal characterization that takes objects, features, and relations as primitives and specifies conceptual domains by combining these primitives in different ways. Explaining how humans are able to learn concepts within all of these domains is a challenge for computational models, but I argue that this challenge can be met by models that rely on a compositional representation language such as predicate logic. The article presents such a model and demonstrates that it accounts well for human concept learning across 11 different domains.

Optimal Behavioral Hierarchy

“…As is clear from the figure, the hierarchy that maximizes the efficiency of representation also maximizes the efficiency of learning. This is no coincidence: It is a well established result from learning theory, echoed in empirical observations of human behavior, that ease of learning is directly related to descriptive complexity [27], [28]. Indeed, this connection has inspired previous efforts to identify useful subtask representations through data compression [21], [29]–[31].…”

Is it better to select or to receive? Learning via active and passive hypothesis testing.

“…As mentioned earlier, in prior work with these types of tasks, leamers appear to rely initially on the explicit system, with a bias toward using unidimensional rules for classification. This would be consistent with a computational level inductive bias toward simpler rule structures (Feldman, 2000;Kemp, 2012). Our prior captures this preference by assigning higher a priori likelihood to mies that involve attention to a single dimension.…”

Dynamic inference and everyday conditional reasoning in the new paradigm

“…This approach is also adopted in other areas of Bayesian Cognitive Science (e.g., Goodman, Ullman, & Tenenbaum, 2011;Kemp, 2012) and the development of general programming languages like CHURCH (Goodman, Mansinghka, Roy, Bonawitz, & Tenenbaum, 2008) which incorporate probabilistic primitive operations is consistent with this view. Such approaches are not inimical to the proposal that people represent probabilistic information in terms of the dependencies they describe as in Causal Bayes Nets (CBNs).…”