Neural Spatial Interaction Predictors and Pattern Detectors

点击翻译按钮获取中文摘要

9300633 GOPAL A form of non-programmed adaptive information processing called Artificial Neural Networks (ANNs) provides an alternative means for performing complicated and creative tasks like pattern classification and completion as well as image and speech recognition. ANNs are inherently parallel in structure, they distribute knowledge throughout the system rather than concentrating it in specific locations, and they are fault- tolerant, learning from and making decisions based on incomplete, noisy, or imprecise information. Because of these characteristics, ANNs offer great potential for future use in processing of geographic information, but that potential has not been widely explored by geographers and regional scientists. Working with an Austrian collaborator, an American geographer will explore a number of potential uses for ANNs in geographic research. Attention in this project will focus on applications to the study of spatial interaction. Application-dependent neural spatial interaction predictors will be designed, implemented, and tested, as will unsupervised ANN-based interaction pattern detectors. ANN approaches to these problems will be compared with traditional statistical approaches through empirical testing with data on Austrian inter-regional telephone calls. This project constitutes an exploratory look into the potential for using ANN approaches in the analysis of geographical patterns and processes. By focusing on spatial interaction, the investigators will explore new analytical perspectives on an important topic for which conventional approaches are well established. Regardless of the outcomes of these comparisons, scientists in geography, regional science, computer science, and related fields will benefit from this research on an innovative new approach to problem solving. ***