Prediction and Control in an Active Environment
The definition of a mechanism which learns to predict the consequences of events and actions in an environment is in progress. Salient features of this research include an attempt to separate prior assumptions (bias) from the learning algorithm proper and, as far as possible, to make them explicit. The emphasis is on events rather than objects — the input from the environment is is not restricted to be a succession of object and relation oriented descriptions of environmental state. Of particular interest is the acquisition of such object based representations. The task domain (learning to predict in a particular class of environments) is defined with these aims in mind.
KeywordsCellular Automaton Event History Label Graph Prior Assumption Neighbourhood Relation
Unable to display preview. Download preview PDF.