|
Discovering Systems of Concepts
Patrick Shafto, Chalres Kemp, Vikash Mansignhka, Matthew Gordon &
Joshua B. Tenenbaum
Abstract
Most natural domains may be represented in multiple ways: animals may
be thought of in terms of their taxonomic groupings or their ecological
niches; foods may be thought of in terms of their nutritional content
or social role. Here we present a computational framework for learning
multiple systems of concepts that capture the structure of a domain of
objects and their properties. We focus on the special case of discovering
multiple ways to categorize objects, such that each system of categories
accounts for a distinct and coherent subset of the objects' features.
A first experiment shows that our {\em CrossCat} model predicts human
learning in an artificial category learning task. A second experiment
shows that the model discovers important structure in two real-world domains.
Traditional models of categorization usually search for a single system
of categories: we suggest that these models do not predict human performance
in our task, and miss important structure in our real world examples.
|
|
