Patent US5408585A successfully refreshed from Google Patents.
US5408585A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Internal connection method for neural networks
Assignee: Thales SA
•
Filing: 1994-07-14
•
Granted: 1995-04-18
•
Expires: 2014-07-14
Abstract
Abstract A method of internal connection for neural network linking successive neuron layers and useful in image analysis and image and signal processing is provided. The neuron output states of a new neuron layer are represented by functions obtained by the method. A weighted summation of functions which represent the output state of a neuron from a preceding layer is carried out. A saturation function is applied to the result of the weighted summation. A function is outputted which is representative of the output state of the neuron layer.
Inventors
- 👤 Gilles Burel
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-07 22:00
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063; G06N3/065
Claims Analysis
Total Claims: 12
Claim 1
Independent Claim
128 words
An internal connection method for neural networks, comprising linking successive neuron layers, each layer having at least one input and at least one output, wherein said neuron layers are connected so that the output of one neuron layer is connected to the input of a successive neuron layer, each neuron layer receiving at said inputs, output functions from a preceding neuron layer, using a weighted summer to carry out a weighted summation of said output functions which represent an output state from the preceding neuron layer, using a saturation function stage means to apply a saturation function to said weighted summation, and using a distribution function stage to apply a distribution function to said saturated summation and outputting a function representative of output states of said neuron layer.
Claim 2
Dependent Claim
128 words
An internal connection method for neural networks, comprising linking successive neuron layers, each layer having at least one input and at least one output, wherein said neuron layers are connected so that the output of one neuron layer is connected to the input of a successive neuron layer, each neuron layer receiving at said inputs, output functions from a preceding neuron layer, using a weighted summer to carry out a weighted summation of said output functions which represent an output state from the preceding neuron layer, using a saturation function stage means to apply a saturation function to said weighted summation, and using a distribution function stage to apply a distribution function to said saturated summation and outputting a function representative of output states of said neuron layer.
Claim 3
Dependent Claim
45 words
The method according to claim 1, further comprising said distribution function stage transforming each weighted summation obtained by said saturation function stage into a function that is a convolution product of a smoothing filter function by a Dirac delta function centered on said weighted summation.
Claim 4
Dependent Claim
16 words
The method according to claim 2, wherein said smoothing filter function is substantially a Gaussian function.
Claim 5
Dependent Claim
19 words
The method according to claim 1, further comprising representing the weights of said weighted summation by a sampled function.
Claim 6
Dependent Claim
28 words
The method of claim 4, further comprising setting sampled weights of said summation function that are outside an interval enclosing asymptotic values of the saturating function to zero.
Claim 7
Dependent Claim
14 words
A method according to claim 5, wherein said saturating function is a hyperbolic tangent.
Claim 8
Dependent Claim
13 words
The method of claim 1, wherein said saturating function is a hyperbolic tangent.
Claim 9
Dependent Claim
17 words
The method of claim 1, further comprising sampling said function representing the output state of said neuron.
Claim 10
Dependent Claim
22 words
The method according to claim 1, further comprising setting up the neural net connections during a learning phase employing a gradient algorithm.
Claim 11
Dependent Claim
109 words
A neural network comprising, a plurality of neurons arranged in layers, each neuron layer having inputs and outputs wherein each neuron layer receives output states as functions from a preceding neuron layer, and wherein each neuron layer comprises a weighted summer for obtaining a weighted summation of functions which represent said output states from the preceding neuron layer, a saturation function stage for obtaining a saturated summation by applying a saturating function to said weighted summation, and a distribution function stage for applying a distribution function to said saturated summation, wherein said distribution stage has an output which outputs a function representative of output states of said neuron layer.
Claim 12
Dependent Claim
109 words
A neural network comprising, a plurality of neurons arranged in layers, each neuron layer having inputs and outputs wherein each neuron layer receives output states as functions from a preceding neuron layer, and wherein each neuron layer comprises a weighted summer for obtaining a weighted summation of functions which represent said output states from the preceding neuron layer, a saturation function stage for obtaining a saturated summation by applying a saturating function to said weighted summation, and a distribution function stage for applying a distribution function to said saturated summation, wherein said distribution stage has an output which outputs a function representative of output states of said neuron layer.
References Cited (Backward Citations)
US4315319A
Cited
US4991092A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1998-04-17 | 1998-10-17 | 1999-04-18 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2002-04-17 | 2002-10-17 | 2003-04-18 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2006-04-17 | 2006-10-17 | 2007-04-18 | $7700.00 | Window Closed / Lapsed |