Adaptive computing system capable of being trained to recognize patterns

• US 3,267,431 A
• Filed: 04/29/1963
• Issued: 08/16/1966
• Est. Priority Date: 04/29/1963
• Status: Expired due to Term

• Current Assignee: Alan G. Konheim, Herbert J. Greenberg
• Original Assignee: Alan G. Konheim, Herbert J. Greenberg
• Inventors: Greenberg, Herbert J., Konheim, Alan G.
• Application Number:
  US04/276,612
• Time in Patent Office: 1,205 Days
• US Class Current: 382/159
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning

Claims

1. AN ADAPTIVE RECOGNITION SYSTEM FOR IDENTIFYING A SPECIMEN BASED ON ITS SIMILARITY OF MEMBERS OF A SET OF REFERENCE PATTERNS COMPRISING, IN COMBINATION:

MEANS FOR GENERATING NONRANDOM FUNCTIONS OF THE SPECIMENS;

MEANS FOR LINEARLY COMPRSING THE GENERATED FUNCTIONS WITH REFERENCE FUNCTIONS TO PROVIDE AN INDICATION OF THE REFERENCE PATTERN THAT IS THE MOST SIMILAR TO THE SPECIMEN FOR EACH SPECIMEN, WHERE EACH REFERENCE FUNCTION CORRESPONDS TO A MEMBER OF THE SET OF REFERENCE PATTERNS;

AND SELECTIVELY-CONTROLLABLE MEANS RESPONSIVE TO THE COMPARISON MEANS FOR ALTERING THE REFERENCE FUNCTIONS DURING A TRAINING MODE OF OPERATION TO IMPROVE THE COMPARISON BETWEEN THE GERATED FUNCTIONS AND A CORRESPONDING REFERENCE FUNCTION THAT IS MOST SIMILAR, AND TO LESSEN THE COMPARISON BETWEEN THE GENERATED FUNCTIONS AND AT LEAST ONE OTHER NONCORRESPONDING REFERENCE FUNCTION AND FOR PROVIDING AN INDICATION OF THE IDENTITY OF THE SPECIMENS BASED ON THEIR SIMILARITY TO THE MEMBERRS OF THE SET OF REFERENCE PATTERNS, AS DETERMINED BY BY THE COMPARISON MEANS, DURING AN OPERATING MODE OF OPERATION.