Method of a machine that learns

• US 5,577,167 A
• Filed: 04/03/1995
• Issued: 11/19/1996
• Est. Priority Date: 11/22/1993
• Status: Expired due to Fees

• Current Assignee: Robert A Brown
• Original Assignee: Robert A Brown
• Inventors: Brown, Robert A.
• Primary Examiner(s): Davis, George B.
• Application Number:
  US08/415,296
• Time in Patent Office: 596 Days
• Field of Search: 395/23, 395/20, 395/21, 395/24, 395/11
• US Class Current: 706/14
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/063 using electronic means

Abstract

A self-learning machine (10) with one or more sensor(s) (110) and actuator(s) (135), and a plurality of memory cells (20) that select actions for the actuator(s) (135) to be carried out during specific sensing and action periods, the selection being determined by the stored values in the memory cells (20) and the conditions identified by its sensor(s) (110) in these sensing and action periods. The stored values are reduced in the selecting memory cells (20) when they select specific actions, and the stored values are increased in selected memory cells (20) according to what actions actually occur within these specific sensing and action periods.

Claims

1. A method of selecting a value of at least one of an actuator means in response to an extant value of at least one of a sensor means in a sensing and action period, said selected value being determined by a record of a historical probability of each of an actual value of said actuator means occurring with each of said extant value of said sensor means in each of said sensing and action period, said method comprising:

a. determining a beginning and an end of said sensing and action period, and

b. identifying said extant value of each said sensor means near a beginning of said sensing and action period, and

c. examining said record of said historical probability belonging to each said extant value of said sensor means in said sensing and action period, and

d. selecting said selected value of each said actuator means, said selected value having a highest of said historical probability of occurring with said extant value of each said sensor means in said sensing and action period, and

e. making an attempt to produce said selected value of each said actuator means within said sensing and action period, and

f. restraining said actuator means except when said attempt to produce said selected value is made, and

g. making a measurement of an actual value of each said actuator means after a predetermined delay from said attempt to produce said selected value of said actuator means, said measurement being made within said sensing and action period, and

h. establishing said record of said historical probability, said record being made according to each said actual value of said actuator means occurring with each said extant value of said sensor means during each of said sensing and action period, whereby

i. said method of selecting a value of said actuator means in response to said value of said sensor means and said record of said historical probability provides a simple self-learning machine.

2. The method of claim 1, more specifically of applying to one of said sensor means and one of said actuator means, said method comprising:

a. making a measurement of a measured value of said sensor means, said measured value corresponding to an extant value of said sensor means at a beginning of said sensing and action period, and

b. maintaining said measured value of said sensor means for a duration of said sensing and action period, and prohibiting a different value of said sensor means from being measured during said duration of said sensing and action period, and

c. identifying an input set of a memory cell means, where said input set belongs to said measured value of said sensor means, and

d. examining a sensitivity of each said memory cell means in said input set, and

e. determining a selecting memory cell means from within said input set, said selecting memory cell having a highest said sensitivity within said input set, and where said selecting memory cell means belongs to a selected value of said actuator means, and

f. making a reduction in said sensitivity of said selecting memory cell means, said reduction being equal to said sensitivity of said selecting memory cell means multiplied by a predetermined constant, said predetermined constant being in a range greater than zero to one, and

g. making an attempt to produce said selected value of said actuator means, and

h. restraining said actuator means except when said attempt to produce said selected value is made, and

i. making a measurement of a measured value of said actuator means, said measured value of said actuator means corresponding to an actual value of said actuator means, said measurement being made after an action period, said action period being long enough to allow said actuator means to attempt to arrive at said selected value, and

j. maintaining said measured value of said actuator means for a remainder of said sensing and action period, and prohibiting a different value of said actuator means from being measured during said remainder of said sensing and action period, and

k. identifying a feedback set of memory cell means, where said feedback set belongs to said measured value of said actuator means, and

l. identifying a feedback selected memory cell means, where said feedback selected memory cell means is at an intersection of said feedback set of memory cells means and said input set of memory cells means, and

m. making an increase in a sensitivity of said feedback selected memory cell means, said increase being equal to a difference between a maximum sensitivity of said feedback selected memory cell means and an extant sensitivity of said feedback selected memory cell means, said difference being multiplied by said predetermined constant, wherein

n. each said selected value of said actuator means has a highest historical probability of occurring with each said measured value of said sensor means within each said sensing and action period, whereby

o. said method provides a simple self-learning machine with said one said sensor means and said one said actuator means.

3. The method of claim 1, more specifically applying to one of said sensor means and a plurality of said actuator means, said method comprising:

a. making a measurement of a measured value of said sensor means, said measured value corresponding to an extant value of said sensor means at a beginning of said sensing and action period, and

b. maintaining said measured value of said sensor means for a duration of said sensing and action period, and prohibiting a different value of said sensor means from being measured during said duration of said sensing and action period, and

c. identifying a plurality of an input set of a memory cell means, where each said input set belongs to said measured value of said sensor means, and wherein each said input set also belongs to each said actuator means uniquely, and

d. examining a sensitivity of each said memory cell means within each said input set, and

e. determining a selecting memory cell means within each said input set, wherein each said selecting memory cell means has a highest said sensitivity within each said input set, and each said selecting memory cell means belongs to each of a selected value of said actuator means, and

f. making a reduction of each said sensitivity of each said selecting memory cell means, said reduction being equal to an extant sensitivity of each said selecting memory cell means multiplied by a predetermined constant, said predetermined constant being in a range greater than zero to one, and

g. making an attempt to produce each said selected value of each of said actuator means, and

h. restraining each said actuator means except when said attempt to produce said selected value is made, and

i. making a measurement of a measured value of each said actuator means, said measured value of each said actuator means corresponding to an actual value of each said actuator means, said measurement being made after an action period, said action period being long enough to allow said actuator means to attempt to arrive at said selected value, and

j. maintaining each said measured value for a remainder of said sensing and action period, and prohibiting a different value of said actuator means from being measured during said remainder of said sensing and action period, and

k. identifying a plurality of a feedback set of said memory cell means, where each said feedback set belongs to each said measured value of each said actuator means uniquely, and

l. identifying a plurality of a feedback selected memory cell means, where each said feedback selected memory cell means is at an intersection of each said feedback set of memory cells means and each said input set of memory cells means, and

m. making an increase in said sensitivity of each said feedback selected memory cell means, said increase being equal to a difference between a maximum sensitivity of each said feedback selected memory cell means and an extant sensitivity of each said feedback selected memory cell means, said difference being multiplied by said predetermined constant, wherein

n. each said selected value of said actuator means has a highest historical probability of occurring with each said measured value of said sensor means within each said sensing and action periods, whereby

o. said method provides a simple self-learning machine with said one said sensor means and said plurality of said actuator means.

4. The method of claim 1, more specifically applying to a plurality of said sensor means and one of said actuator means, said method comprising:

a. making a measurement of a measured value of each said sensor means, said measured value corresponding to an extant value of each said sensor means at a beginning of said sensing and action period, and

b. maintaining each said measured value of each said sensor means for a duration of said sensing and action period, and prohibiting a different value of each said sensor means from being measured during said sensing and action period, and

c. identifying a plurality of an input set of memory cell means, where each said input set belongs to each said measured value of each said sensor means uniquely, and

d. examining a sensitivity of each of said memory cell means in each said input set, and

e. determining a selecting memory cell means within each said input set, said selecting memory cell means forming a selecting set of memory cell means, where said selecting set belongs to a selected value of said actuator means, wherein

f. a lowest sensitivity memory cell means within said selecting set has a higher said sensitivity than a lowest sensitivity memory cell means in any other of an unselecting set of said memory cell means, where each said unselecting set belongs to each of an unselected value of said actuator means and said input set, and

g. reducing each said sensitivity of each said selecting memory cell means, each said reduction being equal to an extant sensitivity of each said selecting memory cell means multiplied by a predetermined constant, and

h. making an attempt to produce said selected value of said actuator means, and

i. restraining said actuator means except when said attempt to produce said selected value is made, and

j. making a measurement of a measured value of said actuator means, said measured value of said actuator means corresponding to an actual value of said actuator means, said measurement being made after an action period, said action period being long enough to allow said actuator means to attempt to arrive at said selected value, and

k. maintaining said measured value for a remainder of said sensing and action period, and prohibiting a different value of said actuator means from being measured during said remainder of said sensing and action period, and

l. identifying a feedback set of memory cell means, where said feedback set of memory cell means belongs to said measured value of said actuator means, and

m. identifying a plurality of a feedback selected memory cell means, where each said feedback selected memory cell means is at an intersection of said feedback set of memory cells means and each said input set uniquely, and

n. making an increase in said sensitivity of each said feedback selected memory cell means, said increase being equal to a difference between a maximum sensitivity of each said feedback selected memory cell means and an extant sensitivity of each said feedback selected memory cell means, said difference being multiplied by said predetermined constant, wherein

o. said selected value of said actuator means has a highest historical probability of occurring with each said measured value of each said sensor means within each said sensing and action periods, whereby

p. said method provides a simple self-learning machine with said plurality of said sensor means and said one said actuator means.

5. The method of claim 1, more specifically applying to a plurality of said sensor means and a plurality of said actuator means, said method comprising:

a. making a measurement of a measured value of each said sensor means, each said measured value corresponding to an extant value of each said sensor means at a beginning of said sensing and action period, and

b. maintaining each said measured value of each said sensor means for a duration of said sensing and action period, and prohibiting a different value of each said sensor means from being measured during said sensing and action period, and

c. identifying a plurality of an input set of memory cell means, where each said input set belongs to each said measured value of each said sensor means uniquely, and where each input set also belongs to each said actuator means uniquely, and

d. examining a sensitivity of each of memory cell means in each said input set of memory cell means, and

e. determining a selecting memory cell means within each said input set, said selecting memory cell means forming a selecting set of memory cell means, where each said selecting set belongs to a selected value of each said actuator means uniquely, wherein

f. a lowest sensitivity memory cell means within each said selecting set has a higher said sensitivity than a lowest sensitivity memory cell means in any other of an unselecting set of said memory cell means, where each said unselecting set belongs to each of an unselected value of each said actuator means and said input set uniquely, and

g. reducing said sensitivity of each said selecting memory cell means, said reduction being equal to an extant sensitivity of each said selecting memory cell means multiplied by a predetermined constant, and

h. making an attempt to produce said selected value of each said actuator means, and

i. restraining each said actuator means except when said attempt to produce said selected value is made, and

j. making a measurement of a measured value of each said actuator means, each said measured value of each said actuator means corresponding to an actual value of each said actuator means, said measurement being made after an action period, said action period being long enough to allow each said actuator means to attempt to arrive at said selected value, and

k. maintaining each said measured value for a remainder of said sensing and action period, and prohibiting a different value of each said actuator means from being measured during said remainder of said sensing and action period, and

l. identifying a plurality of a feedback set of memory cell means, where each said feedback set of memory cell means belongs to each said measured value of each said actuator means, and

m. identifying a plurality of a feedback selected memory cell means, where each said feedback selected memory cell means is at an intersection of said feedback set of memory cells means and said input set of memory cells means, and

n. making an increase in said sensitivity of each said feedback selected memory cell means, said increase being equal to a difference between a maximum sensitivity of each said feedback selected memory cell means and an extant sensitivity of each said feedback selected memory cell means, said difference being multiplied by said predetermined constant, wherein

o. each said selected value of each said actuator means has a highest historical probability of occurring with each said measured value of each said sensor means within each said sensing and action periods, whereby

p. said method provides a simple self-learning machine with said plurality of said sensor means and said plurality of said actuator means.

6. The method of claim 5 further including a method of digitization useful in a digitized self-learning machine, said method comprising:

a. decomposing said extant value of said sensor means into a unique combination of values of a plurality of an aggregate matrix input variable means, said decomposing being accomplished by an encoding means, and

b. selecting a specific combination of values of a plurality of an aggregate output variable means for each said unique combination of values of said plurality of said aggregate input variable means, wherein

c. said specific combination of values is selected according to a highest of said historical probability of a unique combination of values of a plurality of an aggregate output co-variable means occurring with each said unique combination of values of said plurality of aggregate input variable means in said sensing and action period, said historical probability being stored in said memory cell means, and

d. making a synthesis of a unique of a selected value of said actuator means from each said specific combination of values of said plurality of aggregate matrix output variable means, said synthesis being made by a decoding means, and

e. making an attempt to produce said selected value of said actuator means, and

f. restraining said actuator means except when said decoding means makes said attempt to produce said selected value, and

g. making a measurement of a measured value of said actuator means, said measured value of said actuator means corresponding to said actual value of said actuator means, said measurement being made after an action period, said action period being long enough to allow said actuator means to attempt to arrive at said selected value, and

h. decomposing said measured value of said actuator means into said unique combination of values of said plurality of said aggregate matrix output co-variable means, said decomposing being done by an encoding means, and

i. establishing said historical probability between said unique combination of values of said aggregate output co-variables and said unique combination of values of said plurality of said aggregate matrix input variable means, whereby

j. said method provides a digitized self-learning machine, said digitized self-learning machine requiring fewer said memory cell means than an undigitized self-learning machine.