US5257342A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Learning method for a data processing system with neighborhoods
Assignee: Yozan Inc
•
Filing: 1991-04-03
•
Granted: 1993-10-26
•
Expires: 2011-04-03
Abstract
Abstract A learning method for a neural network type data processing system determines activation patterns in an input layer and output layer arbitrarily, increases weights of synapses in a middle layer and the output layer so that neuron activate with more than a certain rate among those corresponding to neurons in the input layer and the output layer and repeats the same process for each neuron in the middle layer. The input layer and output layer possess a plurality of neurons which activate and output certain data according to a specific result and the middle layer is between the input layer and output layer. The middle layer also possesses a plurality of neurons which are connected to each neuron in the input layer and output layer.
Inventors
- 👤 Sunao Takatori
- 👤 Makoto Yamamoto
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-02 15:16
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/08; G06N3/082; G06N3/04; G06V; G06V10/00; G06V10/70; G06V10/82; G06V30/00; G06V30/10; G06V30/18; G06V30/1801; G06V30/18019; G06V30/18038; G06V30/18048; G06V30/18057; G06V30/19; G06V30/191; Y; Y10; Y10S; Y10S706/00; Y10S706/90
Claims Analysis
Total Claims: 7
Claim 1
Independent Claim
279 words
A learning method for transforming an untrained network to a trained network, said network comprising an input layer, a middle layer and an output layer, each said layer including a plurality of neurons, each said neuron being a signal processing element, said middle layer being arranged between said input and output layers wherein each neuron in said middle layer is connected to each neuron in said input and output layers, said method comprising the steps of: determining an activation pattern of said neurons in said input layer; determining an activation pattern of said neurons in said output layer; defining a neighborhood of said neurons in said input and output layers corresponding to a neuron in said middle layer; determining a rate of activation in said neighborhood of neurons, said rate being the total number of neurons activating in said neighborhood in said input and output layers divided by the total number of neurons in said neighborhood in said input and output layers; comparing said rate to a threshold value; determining that said neuron in said middle layer has a tendency to activate when said rate is greater than said threshold value; increasing the value of weights applied to all neurons in said input layer which are connected to said neuron in said middle layer and weights applied to all neurons in said output layer which are connected to said neuron in said middle layer, in order to increase the likelihood of activation of said neuron in said middle layer; and repeating the above steps for each said neuron in said middle layer, thereby generating a desired activation pattern of neurons in said middle layer corresponding to said trained network.
Claim 2
Dependent Claim
279 words
A learning method for transforming an untrained network to a trained network, said network comprising an input layer, a middle layer and an output layer, each said layer including a plurality of neurons, each said neuron being a signal processing element, said middle layer being arranged between said input and output layers wherein each neuron in said middle layer is connected to each neuron in said input and output layers, said method comprising the steps of: determining an activation pattern of said neurons in said input layer; determining an activation pattern of said neurons in said output layer; defining a neighborhood of said neurons in said input and output layers corresponding to a neuron in said middle layer; determining a rate of activation in said neighborhood of neurons, said rate being the total number of neurons activating in said neighborhood in said input and output layers divided by the total number of neurons in said neighborhood in said input and output layers; comparing said rate to a threshold value; determining that said neuron in said middle layer has a tendency to activate when said rate is greater than said threshold value; increasing the value of weights applied to all neurons in said input layer which are connected to said neuron in said middle layer and weights applied to all neurons in said output layer which are connected to said neuron in said middle layer, in order to increase the likelihood of activation of said neuron in said middle layer; and repeating the above steps for each said neuron in said middle layer, thereby generating a desired activation pattern of neurons in said middle layer corresponding to said trained network.
Claim 3
Dependent Claim
156 words
The learning method of claim 1 comprising the steps of: a. increasing the value of said threshold so that less neurons in said middle layer have a tendency to activate; b. comparing said rate in said neighborhood of neurons to said increased threshold value; c. determining that said neuron in said middle layer has a tendency to activate when said rate is greater than said increased threshold value; d. increasing the value of weights applied to all neurons in said input layer which are connected to said neurons in said middle layer and weights applied to all neurons in said output layer when are connected to said neuron in said middle layer, in order to increase the likelihood of activation of said neuron in said middle layer; and e. repeating steps b.-d. for each neuron in said middle layer, thereby generating a desired activation pattern of neurons in said middle layer corresponding to said trained network.
Claim 4
Dependent Claim
221 words
The learning method of claim 2 further including the steps of: a. adding an additional middle layer between said input layer and said middle layer or between said middle layer and said output layer, said additional middle layer having a plurality of neurons; b. connecting each neuron in said additional middle layer to each neuron in the two layers adjacent to said additional middle layer; c. defining a neighborhood of neurons in said adjacent layers corresponding to a neuron in said additional middle layer; d. determining a rate of activation in said neighborhood of neurons, said rate being the total number of neurons activating in said neighborhood divided by the total number of neurons in said neighborhood; e. comparing said rate to a threshold value; f. increasing the value of weights applied to all neurons in one of said adjacent layers which are connected to said neuron in said additional middle layer and weights applied to all neurons in said other adjacent layer which are connected to said neuron in said additional middle layer, in order to increase the likelihood of activation of said neuron in said additional middle layer; and d. repeating steps c.-f. for each neuron in said additional middle layer, thereby generating a desired activation pattern of neurons in said additional middle layer corresponding to said trained network.
Claim 5
Dependent Claim
167 words
The learning method of claim 3 further comprising the steps of: a. increasing the value of said threshold so that less neurons in said additional middle layer have a tendency to activate; b. comparing said rate in said neighborhood of neurons to said increased threshold value; c. determining that said neuron in said additional middle layer has a tendency to activate when said rate is greater than said increased threshold value; d. increasing the value of weights applied to all neurons in one of said adjacent layers which are connected to said neuron in said additional middle layer and weights applied to all neurons in said other adjacent layer which are connected to said neuron in said additional middle layer, in order to increase the likelihood of activation of said neuron in said additional middle layer; and e. repeating steps b.-d. for each neuron in said additional middle layer, thereby generating a desired activation pattern of neurons in said additional middle layer corresponding to said trained network.
Claim 6
Dependent Claim
50 words
The learning method of claim 1 wherein said neighborhood includes a neuron in said input layer and a neuron in said output layer corresponding to the same position as said neuron in said middle layer and all neurons adjacent to said neuron in said input layer and said output layer.
Claim 7
Dependent Claim
45 words
The learning method of claim 1, wherein said learning method is a method for recognizing characters, said activation pattern in said input layer corresponds to characteristics data of an image, and said activation pattern in said output layer corresponds to character data of said image.
References Cited (Backward Citations)
US4874963A
Cited
US5033066A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1996-10-25 | 1997-04-26 | 1997-10-26 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2000-10-25 | 2001-04-26 | 2001-10-26 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2004-10-25 | 2005-04-26 | 2005-10-26 | $7700.00 | Window Closed / Lapsed |