GOAL-DIRECTED PATTERN RECOGNITION

Publisher Summary In theoretical investigations of pattern recognition, it is common practice to represent the system-environment context minimally, usually by a partitioned set indicating the pattern categories. While this can be fruitful, as it has been for some kinds of character recognition, both thought and practice point up its limitations. It is particularly inadequate in goal-directed situations where the patterns to be recognized depend heavily on those features of the environment relevant to attainment of the system's goals. The object of a goal-directed pattern recognition system is to discover, consistently recognize, and utilize critical features of any environmental situation within its competence. This chapter highlights the problems entailed in constructing such systems and with ways of resolving these problems. It explores relevant aspects of Samuel's well-known work, with the intent of exhibiting some features of goal-directed pattern recognition in a familiar context.