Pattern association central subsystem and a perception learning system

• US 5,321,795 A
• Filed: 05/24/1991
• Issued: 06/14/1994
• Est. Priority Date: 05/24/1991
• Status: Expired due to Term

• Current Assignee: Santiago Alvarez De Toledo
• Original Assignee: Santiago Alvarez De Toledo
• Inventors: Alvarez de Toledo, Santiago
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): DAVIS, GEORGE B
• Application Number:
  US07/705,300
• Time in Patent Office: 1,117 Days
• Field of Search: 395/10, 395/902, 395/903, 395/904, 395/906, 395/21, 395/22, 395/11, 395/24
• US Class Current: 706/62
• CPC Class Codes: B25J 9/161 Hardware, e.g. neural networks, fuzzy logic, interfaces, processor
• CPC Class Codes: G05B 21/02 electric
• CPC Class Codes: G05B 2219/33056 Reinforcement learning, agent acts, receives reward, emotion, action selective
• CPC Class Codes: G05B 2219/39271 Ann artificial neural network, ffw-nn, feedforward neural network
• CPC Class Codes: G05B 2219/40115 Translate goal to task program, use of expert system
• CPC Class Codes: G05D 1/0221 involving a learning process
• CPC Class Codes: G05D 1/0231 using optical position detecting means position-fixing by using electromagnetic waves other than radio waves, e.g. optical position detecting means G01S5/16
• CPC Class Codes: G06F 15/8038 Associative processors

Abstract

A general purpose acting intelligent system includes a pattern association central subsystem which modifies actions taken by an output subsystem through a statistical valuation of the positive or negative effects of associations based upon the proximity in time between perception patterns developed from responses by sensors in an input subsystem, and actions previously taken by the output subsystem in response to such perception patterns. The pattern association central subsystem also modifies output actions taken based upon statistical associations between perception patterns perceived by the input subsection and positive or negative results detected by the input subsection to speed up the response of the system.

Claims

1. An acting intelligent system comprising:

input means including input sensors responsive to selected external stimuli to generate input signals, perception means responsive to said input signals to generate perception pattern signals classifying by similarity various input signals and combinations of input signals, and evaluation means for comparing specified external stimuli to selected evaluation criteria to selectively generate positive and negative response signals;

output means for generating selected output actions which directly or indirectly result over time in changes in at least one of said selected and specified external stimuli; and

processing means including means for generating association signals associating specified output actions with said perception pattern signals statistically based upon a repetition and a proximity in time of said perception pattern signals, said selected output actions, and said positive and negative response signals, and means for selecting said selected output actions to be generated by said output means based upon said association signals.

2. The system of claim 1 wherein said means generating said association signals includes means for assigning assigned values to said association signals, said assigned values being positive for an association of a perception pattern signal and an output action with a positive response signal and said assigned value being negative for an association of a perception pattern signal and an output action with a negative response signal, and wherein said means for selecting said selected output actions selects an output action from an association with a positive association value and inhibits selection of an output action from an association with a negative association value.

3. The system of claim 2 wherein said means for assigning values for said association signals includes means for increasing an assigned value in magnitude for each association which is repeated and decreasing in magnitude over time the assigned values of associations not repeated.

4. The system of claim 3 wherein said means for decrease in magnitude over time assigned values which have reached a predetermined level at which the assigned value of such association is maintained.

5. The system of claim 2 wherein said evaluation means includes means for varying a level of the positive and negative response signals based upon an intensity of the specified external stimuli, and wherein said means assigning values to said association signals, assigns said assigned values taking into account the level of the positive and negative response signals, and wherein said means for selecting a selected output actions selects as said selected output action, or inhibits the output action of the association with the greatest magnitude.

6. The system of claim 5 wherein said means for assigning assigned values to said association signals increases in magnitude the assigned value of the association signal for each association which is repeated and decreases in magnitude over time the assigned values of the association signals for associations not repeated.

7. The system of claim 5 wherein said means generating association signals includes means for generating response propagation signals associating perception pattern signals and response signals statistically based upon repetition and time proximity of said perception pattern signals to following response signals, and wherein the means assigning values to said association signals includes means assigning values based also upon said response propagation signals.

8. The system of claim 7 including means for assigning said response propagation signals values which are increased in magnitude for associations between perception pattern signals and following response signals which are repeated and decreased in magnitude over time for associations between perception pattern signals and following response signals which are not repeated, and wherein said means for assigning values to said association signals uses as said level of said response signal the greater of the intensity of said specified external stimuli and the magnitude of said response propagation signals.

9. The system of claim 8 wherein said response signals and response propagation signals may each have simultaneous positive and negative values and wherein the final value of the response signal and response propagation signal from which the output action is selected or inhibited is the algebraic summation of the positive value and the negative value.

10. The system of claim 8 wherein said response propagation signals values which have reached a predetermined level are not decreased in magnitude over time and are maintained at such level.

11. The system of claim 1 wherein said processing means for includes means for generating response propagation signals associating perception pattern signals and response signals statistically based upon repetition and time proximity of said perception pattern signals to following response signals, and wherein said means for selecting or inhibiting said selected output action includes means for selecting output actions based also upon said response propagation signals.

12. The system of claim 1- wherein said means for selecting said selected output actions includes means for selecting a random output action as a selected output action to be generated by said output means in response to a perception pattern signal generated by said input means for which no association signal associating that perception pattern signal with an output action has been generated.

13. The system of claim 1 wherein said perception means comprise one of a neural network and an expert system.

14. The system of claim 13 wherein said neural networks comprise unsupervised neural networks.

15. The system of claim 1 wherein said output means comprises one of a neural network and an expert system dedicated to generate output action patterns classifying by similarity various output actions and combinations of output actions.

16. The system of claim 15 wherein said neural network comprises an unsupervised neural network.

17. An acting intelligent system comprising:

input means including input sensors responsive to selected external stimuli to generate input signals, perception means responsive to said input signals to generate perception pattern signals classifying by similarity various input signals and combinations of input signals, and evaluation means comparing specified external stimuli to selected evaluation criteria to selectively generate positive and negative response signals;

output means generating selected output actions which directly or indirectly result over time in changes in at least one of said selected and specified external stimuli; and

processing means generating and storing association signals associating output actions, perception pattern signals and response signals, said processing means strengthening association signals for associations which repetitively produce positive response signals and selecting as a selected output action to be generated by the output means an output action with the strongest association to current perception pattern signals.

18. The system of claim 17 wherein said processing means includes means for weakening over time association signals for associations which are not repeated.

19. The system of claim 17 wherein said processing means includes means for selecting as said selected output action a random action when no association signal has been established for an association between current perception pattern signals and an output action.

20. The system of claim 17 wherein said processing means includes means for generating response propagation signals associating perception pattern signals and response signals, said processing means strengthening said response propagation signals for associations between perception pattern signals and response signals which repetitively include a positive response signal, and including said response propagation signals in selecting said selected output actions.