Combining Low-Level Perception and Expectations in Conceptual Learning
Abstract
This report results from a contract tasking University of Nottingham as follows: The contractor will investigate the interplay between low-level feature extraction and high-level concept formation, using "chunking mechanisms." The model will be presented with pictures of various objects, and will extract the relevant features before using them to build up a discrimination network for classifying novel pictures. The high level chunks acquired by the model will be used to direct eye movements, which in turn will determine what will be learned. Questions of interest will include how robust the model is to current expectations, noisy input (to simulate poor visibility) and varying object orientation. The behavior of the model will be compared to human behavior, e.g., learning to categorize images of planes, object identification from aerial photographs, etc. A successfully implemented model will help predict factors which affect human performance in visual recognition, and suggest support techniques for anticipating and correcting errors.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 12, 2004
- Accession Number
- ADA424888
Entities
People
- Fernand Gobet
- Peter Lane
Organizations
- University of Nottingham