Dealing with large sets of stimuli in learning tasks (abstract only)
Roger Ferguson, William W. McMillan · 1985
Most previous AI research in learning has been aimed at achieving competence in sophisticated logical domains such as mathematical reasoning. Such systems usually handle small problems very well. However, many applications of learning may be required to deal with large numbers of inputs (<1000) that are available simultaneously from sensors and other sources. In addition, responses may have to be selected from a large set. Borrowing ideas from stimulus sampling theory, we are working on programs that will handle problems in this domain. Operations include handling probabilistic relations between inputs and behaviors, changing relations through reinforcement, selectively sampling stimuli, and grouping related stimuli.