Learning Spatial Prototypes

Abstract

This paper considers spatial prototypes as cognitive structures that are encoded into memory through experiences with spatial information. A prototype represents a category of information in your memory and serves as an abstraction of the most typical or central member of that category. Results from an experimental study are compared with results from a simulation. The purpose of the experiment is to understand the factors that influence the formation of spatial prototypes in human memory. The purpose of the simulation is to evaluate the ability of neural network models to learn spatial prototypes. Human subjects learn the nature of four categories of maps by viewing individual maps that are instances of the categories. The prototypes for the categories are defined by the variation of the location, size, and orientation of an island on the maps. Subjects then evaluate all possible maps to determine how typical each map is of the category. An analysis of variance reveals that subjects are able to encode the expected prototypes. The analysis also indicates that the order of the maps in the learning sequences affects how clearly the prototypes are defined. Presenting maps with prototypical characteristics early in the sequence appears to provide the best learning experience. Auto-associator neural networks are provided the same information learned by the human subjects and are able to learn the prototypes better than the human subjects. Geographers may find neural networks useful for modeling many types of visual and behavioral processes where decisions are intuitive rather than logical.

Key Words:

• cognitive structures
• decision making
• learning
• neural networks
• spatial prototypes