Self-organizing circuits

• US 5,515,454 A
• Filed: 07/30/1992
• Issued: 05/07/1996
• Est. Priority Date: 08/06/1981
• Status: Expired due to Fees

• Current Assignee: Buckley B. Shawn
• Original Assignee: Buckley B. Shawn
• Inventors: Buckley, B. Shawn
• Primary Examiner(s): Greening, Wendy R.
• Assistant Examiner(s): CHRISTENSEN, ANDREW BRUCE
• Application Number:
  US07/922,153
• Time in Patent Office: 1,377 Days
• Field of Search: 382/14, 382/15, 382/156, 382/157, 382/158, 395/21-24, 395/26, 395/27, 395/20
• US Class Current: 382/157
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning
• CPC Class Codes: G09B 25/02 of industrial processes; of machinery

Abstract

A self-organizing circuit providing improved performance and reduction in costs. The improvements are of two basic types: those that apply to improved circuit design and those that apply to improved "teaching" of the circuit. A method to allow the circuit elements to learn new patterns quickly is provided. Also, a mechanism by which serial or sequential information can be learned is disclosed. Finally, the invention includes mechanisms by which the circuits can be simplified by reducing the number of interconnections within the circuit. Improved teaching of the circuit includes ways by which the self-organizing circuit can be quickly taught new patterns. First by making each input to a subcircuit compete against the many other inputs to that subcircuit, by weighting each input according to simple Boolean functions, and lastly by incorporating a method by which information can be added to the circuit after the circuit has already learned some information. The circuit makes better distinctions between patterns by incorporating modified subcircuits which are change-sensitive and by making the subcircuit competition be sensitive to change. Lastly, a method of stabilizing and destabilizing subcircuits using signals which are sent to all nodes, lets the subcircuits organize themselves into persistent patterns.

Claims

1. A self-organizing system having a system input signal with constituent elements comprising:

self-organizing means for providing an output signal corresponding to the analysis of the constituent elements of said input signal, said self-organizing means further comprising a plurality of subcircuits, each subcircuit having a changeable state and at least one subcircuit adapted to receive messages of the subcircuit state from at least one other subcircuit, said subcircuits organized into a plurality of levels;

an input set composed of both the constituent elements of said system input signal and the output state of each said subcircuit, each subcircuit adapted to receive input messages from at least one element of the input set and which input messages can originate from other subcircuits in previous levels, in the same level and in higher levels, wherein said input messages provide relative amounts of positive and negative weighting on said subcircuits;

voting means for voting on the state of each said subcircuit based on a comparison of the relative amounts of positive and negative weighting which the input messages have on each subcircuit;

weighting modification means for modifying the weighting of the inputs to each said subcircuit based upon the changes of state of the received input messages.

2. The system of claim 1 wherein said weighting means further comprises universal means for additionally modifying the weighting of the inputs of said subcircuits using universal signals directing said system as a whole between associating said system input with previously learned system inputs and distinguishing said system input from previously learned system inputs.

3. The system of claim 2 wherein said weighting means further comprises comparison means for comparing the relative amount of positive and negative input to said voting means, wherein the comparison in conjunction with said universal means determines whether weight modification of the subcircuit will change from its present value.

4. The system of claim 3 wherein said weighting means has a growth capability which can be limited by the comparison of the relative amounts of positive and negative input provided by said comparison means.

5. The system of claim 1 further comprising time-filtering means for time-filtering said changes of state to exclude from modification subcircuit inputs which change faster than a first determined value and slower than a second determined value.

6. The system of claim 1 in which the changes of subcircuit state include changing from a firing state to a non-firing state and changing from a non-firing state to firing state.

7. The system of claim 1 in which said weighting means is based upon competition among time-filtered Boolean functions computed between the received messages of a subcircuit and the state of that subcircuit and which accumulate over a period of time.

8. The system of claim 2 further comprising destabilizing means, associated with a subcircuit which has changed state, for modifying the weighting of positive and negative inputs to the voting means of a different subcircuit to make that said subcircuit more likely to change state.

9. The system of claim 2 further comprising stabilizing means, associated with a subcircuit which has changed state, for modifying the weighting of positive and negative inputs to the voting means of a different subcircuit to make that said subcircuit less likely to change state.

10. The system of claim 8 wherein said universal means directs all subcircuits which have changed state to use said destabilizing means to propagate changed states to other subcircuits.

11. The system of claim 9 wherein said universal means directs all subcircuits which have changed state to use stabilizing means to ensure that these same subcircuits which have changed state will change in the same way in the future.

12. The system of claim 1 in which the changes of subcircuit state includes only changing from a non-firing state to a firing state.

13. The system of claim 1 in which said changes of state of the received input messages cycle from one state to another state and back again at different cycling rates.