Pattern recognition by means of automatic analogue apparatus

Abstract

The problem of synthesizing apparatus that will automatically simulate man's ability to recognize and to learn to recognize patterns is discussed and it is concluded that analogue circuits, rather than the digital switching circuits that have been employed in the past, provide the simpler solution. A new circuit unit that possesses many of the essential functional characteristics exhibited by nerve cells in the brain is derived from earlier work on the electrical simulation of nervous-system functional activity and forms the basic element of the circuits.The new analogue apparatus consists of a number of distinct functional circuits arranged in a definite sequence, through which signals derived from the patterns to be recognized pass simultaneously on their way to the final output terminals. Classification information may be built into the apparatus initially if it is available, but if not, it can be stored automatically in a special unit during a setting-up procedure in which samples of the pattern types that the apparatus will be required to recognize are presented, together with identification signals. Low-resolution automatic pattern-recognition apparatus is described, and examples illustrate the setting-up procedure and subsequent performance of the apparatus.