Patent FR2550596A1 successfully refreshed from Google Patents.
Patent CH303025A successfully refreshed from Google Patents.
Patent US5239619A successfully refreshed from Google Patents.
US5239619A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Learning method for a data processing system having a multi-layer neural network
Assignee: Yozan Inc
•
Filing: 1991-04-23
•
Granted: 1993-08-24
•
Expires: 2011-04-23
Abstract
Abstract A learning method for a neural network having at least an input neuron layer, an output neuron layer, and a middle neuron layer between the input and output layers. Each of the layers include a plurality of neurons which are coupled to corresponding neurons in adjacent neural layers. The learning method performs a learning function on the neurons of the middle layer on the basis of the respective outputs, or "ignition patterns", of the neurons in the neural layers adjacent to the middle layer. The ignition pattern of neurons in the input layer is decided artificially according to a preferable image pattern to be input. The ignition pattern of neurons in the output layer is decided artificially according to the ignition pattern of the input layer neurons, wherein the ignition pattern of the output layer neurons is predetermined to correspond to a code or pattern preferable for a user. The ignition pattern of the middle layer neurons, coupled to the associated neurons of the respective input and output layers, is then decided according to the ignition pattern of the input layer and the output layer.
Inventors
- 👤 Sunao Takatori
- 👤 Makoto Yamamoto
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-05 17:34
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/08; G06N3/082
Claims Analysis
Total Claims: 19
Claim 1
Independent Claim
313 words
A method for training a data processing system to correctly initiate a course of action based on digital signals supplied by a monitoring system of an environment, the method comprising the steps of: (a) selecting for an input neural layer, a first ignition pattern representative of digital signals supplied by said monitoring system, said input neural layer being part of a neural network in said data processing system, said neural network further including a middle neural layer and an output neural layer, said input neural layer having neurons connected to neurons of said middle neural layer and said neurons of said middle neural layer being connected to neurons of said output neural layer; (b) selecting for said output neural layer, a second ignition pattern representative of output for initiating a course of action in accordance with said first ignition pattern; (c) determining a third ignition pattern for said middle neural layer according to said first ignition pattern and said second ignition pattern so that a correct course of action will be initiated in response to digital signals input from said monitoring system, said middle neural layer being disposed to be responsive to said input neural layer and so that said output neural layer is responsive to said middle neural layer, wherein said determining step comprises for each neuron in said middle layer the steps of, (i) determining whether a set of neurons from said input and output neural layers corresponding to a neuron of said middle neural layer ignite at more than a predetermined rate, a set of corresponding neurons differing for each neuron of said middle layer; and (ii) increasing a weight of the neuron of said middle neural layer and neurons of said output neural layer in said set of corresponding neurons when said determining step (i) determines that said set of corresponding neurons ignite at more than a predetermined rate.
Claim 2
Independent Claim
313 words
A method for training a data processing system to correctly initiate a course of action based on digital signals supplied by a monitoring system of an environment, the method comprising the steps of: (a) selecting for an input neural layer, a first ignition pattern representative of digital signals supplied by said monitoring system, said input neural layer being part of a neural network in said data processing system, said neural network further including a middle neural layer and an output neural layer, said input neural layer having neurons connected to neurons of said middle neural layer and said neurons of said middle neural layer being connected to neurons of said output neural layer; (b) selecting for said output neural layer, a second ignition pattern representative of output for initiating a course of action in accordance with said first ignition pattern; (c) determining a third ignition pattern for said middle neural layer according to said first ignition pattern and said second ignition pattern so that a correct course of action will be initiated in response to digital signals input from said monitoring system, said middle neural layer being disposed to be responsive to said input neural layer and so that said output neural layer is responsive to said middle neural layer, wherein said determining step comprises for each neuron in said middle layer the steps of, (i) determining whether a set of neurons from said input and output neural layers corresponding to a neuron of said middle neural layer ignite at more than a predetermined rate, a set of corresponding neurons differing for each neuron of said middle layer; and (ii) increasing a weight of the neuron of said middle neural layer and neurons of said output neural layer in said set of corresponding neurons when said determining step (i) determines that said set of corresponding neurons ignite at more than a predetermined rate.
Claim 3
Dependent Claim
50 words
A method as recited in claim 1, further comprising the step of: (d) repeating said steps (a), (b) and (c) for each of a plurality of data patterns to be processed, said plurality of data patterns being a representative sample of data to be processed by said data processing system.
Claim 4
Dependent Claim
122 words
A method as recited in claim 1, further comprising the steps of: inserting another middle neural layer between said input neural layer and said middle neural layer after completing said step (c), said another middle neural layer being disposed to be responsive to said input neural layer and so that said middle neural layer is responsive to said another middle neural layer, said neurons of said input neural layer being connected to neurons of said another middle neural layer and said neurons of said another middle neural layer being connected to said neurons of said middle neural layer; and repeating said steps (a) to (c) for said another middle neural layer, said second ignition pattern being selected for said middle neural layer.
Claim 5
Dependent Claim
122 words
A method as recited in claim 1, further comprising the steps of: inserting another middle neural layer between said middle neural layer and said output neural layer after completing said step (c), said another middle neural layer being disposed to be responsive to said middle neural layer and so that said output neural layer is responsive to said another middle neural layer, said neurons of said middle neural layer being connected to neurons of said another middle neural layer and said neurons of said another middle neural layer being connected to said neurons of said output neural layer; and repeating said steps (a) to (c) for said another middle neural layer, said second ignition pattern being selected for said middle neural layer.
Claim 6
Dependent Claim
39 words
A method as recited in claim 1, wherein said selecting steps (a) and (b) select said first and second ignition patterns to have an even distribution of neurons in each of said input, middle and output neural layers firing.
Claim 7
Dependent Claim
14 words
A method as in claim 1, wherein said monitoring system comprises a video camera.
Claim 8
Dependent Claim
18 words
A method as in claim 1, wherein said neurons of input, middle and output layers are operational amplifiers.
Claim 9
Dependent Claim
54 words
A method as in claim 1, wherein the weight of said selected neuron of said middle neural layer and said corresponding neurons of said output neural layer are increased by altering resistance of variable resistors corresponding to said selected neuron of said middle neural layer and said corresponding neurons of said output neural layer.
Claim 10
Dependent Claim
78 words
A learning method as recited in claim 1 , wherein step (i) includes the step of determining a predetermined number of neurons from said input and output neural layers having a spatial relationship to the neuron of said middle neural layer as said set of corresponding neurons, and said predetermined number is less than a total number of neurons in said input neural layer and is less than a total number of neurons in said output neural layer.
Claim 11
Independent Claim
324 words
A method for training a data processing system to correctly initiate a course of action based on digital signals supplied by a monitoring system of an environment, the method comprising the steps of: (a) selecting for an input neural layer, a first ignition pattern representative of digital signals supplied by said monitoring system, said input neural layer being part of a neural network in said data processing system, said neural network further including a middle neural layer and an output neural layer, said input neural layer having neurons connected to neurons of said middle neural layer and said neurons of said middle neural layer being connected to neurons of said output neural layer; (b) selecting for said output neural layer, a second ignition pattern representative of output for initiating a course of action in accordance with said first ignition pattern; (c) determining a third ignition pattern for said middle neural layer according to said first ignition pattern and said second ignition pattern so that a correct course of action will be initiated in response to digital signals input from said monitoring system, said middle neural layer being disposed to be responsive to said input neural layer and so that said output neural layer is responsive to said middle neural layer, wherein said determining step comprises the steps of: (i) selecting a selected neuron of said middle neural layer; (ii) adding output values from neurons from said input neural layer corresponding to said selected neuron of said middle neural layer to a weighted output value from a neuron from said output neural layer corresponding to said selected neuron to obtain an addition result; (iii) increasing a weight of said selected neuron of said middle neural layer and said corresponding neuron of said output neural layer when said addition result from said adding step (ii) exceeds a predetermined value; and (iv) selecting another neuron from said middle neural layer and repeating said steps (i) to (iii).
Claim 12
Independent Claim
324 words
A method for training a data processing system to correctly initiate a course of action based on digital signals supplied by a monitoring system of an environment, the method comprising the steps of: (a) selecting for an input neural layer, a first ignition pattern representative of digital signals supplied by said monitoring system, said input neural layer being part of a neural network in said data processing system, said neural network further including a middle neural layer and an output neural layer, said input neural layer having neurons connected to neurons of said middle neural layer and said neurons of said middle neural layer being connected to neurons of said output neural layer; (b) selecting for said output neural layer, a second ignition pattern representative of output for initiating a course of action in accordance with said first ignition pattern; (c) determining a third ignition pattern for said middle neural layer according to said first ignition pattern and said second ignition pattern so that a correct course of action will be initiated in response to digital signals input from said monitoring system, said middle neural layer being disposed to be responsive to said input neural layer and so that said output neural layer is responsive to said middle neural layer, wherein said determining step comprises the steps of: (i) selecting a selected neuron of said middle neural layer; (ii) adding output values from neurons from said input neural layer corresponding to said selected neuron of said middle neural layer to a weighted output value from a neuron from said output neural layer corresponding to said selected neuron to obtain an addition result; (iii) increasing a weight of said selected neuron of said middle neural layer and said corresponding neuron of said output neural layer when said addition result from said adding step (ii) exceeds a predetermined value; and (iv) selecting another neuron from said middle neural layer and repeating said steps (i) to (iii).
Claim 13
Dependent Claim
50 words
A method as recited in claim 10, further comprising the step of: (d) repeating said steps (a), (b) and (c) for each of a plurality of data patterns to be processed, said plurality of data patterns being a representative sample of data to be processed by said data processing system.
Claim 14
Dependent Claim
122 words
A method as recited in claim 10, further comprising the steps of: inserting another middle neural layer between said input neural layer and said middle neural layer after completing said step (c), said another middle neural layer being disposed to be responsive to said input neural layer and so that said middle neural layer is responsive to said another middle neural layer, said neurons of said input neural layer being connected to neurons of said another middle neural layer and said neurons of said another middle neural layer being connected to said neurons of said middle neural layer; and repeating said steps (a) to (c) for said another middle neural layer, said second ignition pattern being selected for said middle neural layer.
Claim 15
Dependent Claim
122 words
A method as recited in claim 10, further comprising the steps of: inserting another middle neural layer between said middle neural layer and said output neural layer after completing said step (c), said another middle neural layer being disposed to be responsive to said middle neural layer and so that said output neural layer is response to said another middle neural layer, said neurons of said middle neural layer beint connected to neurons of said another middle neural layer and said neurons of said another middle neural layer being connected to said neurons of said output neural layer; and repeating said steps (a) to (c) for said another middle neural layer, said second ignition pattern being selected for said middle neural layer.
Claim 16
Dependent Claim
39 words
A method as recited in claim 10, wherein said selecting steps (a) and (b) select said first and second ignition patterns to have an even distribution of neurons in each of said input, middle and output neural layers firing.
Claim 17
Dependent Claim
14 words
A method as in claim 10, wherein said monitoring system comprises a video camera.
Claim 18
Dependent Claim
18 words
A method as in claim 10, wherein said neurons of input, middle and output layers are operational amplifiers.
Claim 19
Dependent Claim
54 words
A method as in claim 10, wherein the weight of said selected neuron of said middle neural layer and said corresponding neurons of said output neural layer are increased by altering resistances of variable resistors corresponding to said selected neuron of said middle neural layer and said corresponding neurons of said output neural layer.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1996-08-23 | 1997-02-22 | 1997-08-24 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2000-08-23 | 2001-02-22 | 2001-08-24 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2004-08-23 | 2005-02-22 | 2005-08-24 | $7700.00 | Window Closed / Lapsed |