US5664066A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Intelligent system for automatic feature detection and selection or identification
Assignee: United States Department of Energy
•
Filing: 1994-09-09
•
Granted: 1997-09-02
•
Expires: 2014-09-09
Abstract
Abstract A neural network uses a fuzzy membership function, the parameters of which are adaptive during the training process, to parameterize the interconnection weights between an (n-1)'th layer and an n'th layer of the network. Each j'th node in each k'th layer of the network except the input layer produces its output value yk,j according to the function ayer k-1, i indexes the nodes of layer k-1 and all the wk,i,j are interconnection weights. The interconnection weights to all nodes j in the n'th layer are given by wn,i,j=wn,j(i, pn,j,1, . . . , pn,j,pn). The apparatus is trained by setting values for at least one of the parameters pn,j,1, . . . , pn,j,Pn. Preferably the number of parameters Pn is less than the number of nodes Nn-1 in layer n-1. wn,j(i,pn,j,1, . . . , pn,j,Pn) can be convex in i, and it can be bell-shaped. Sample functions for wn,j(i, pn,j,1, . . . , pn,j,Pn) include <IMAGE>
Inventors
- 👤 Chuen-Tsai Sun
- 👤 Jyh-Shing Jang
- 👤 Chi-Yung Fu
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-04 09:37
CPC Classifications
H; H01; H01J; H01J37/00; H01J37/32; H01J37/32917; H01J37/32935; G; G06; G06N; G06N3/00; G06N3/02; G06N3/04; G06N3/043; G06F; G06F2218/00; G06F2218/12
Claims Analysis
Total Claims: 53
Claim 1
Independent Claim
164 words
A neural network having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each k'th layer having N k nodes, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU13## where i indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,i,j =w.sub.i,j (i, p.sub.n,j,1, . . . ,p.sub.n,j,P.sbsb.n,j), each of said P n ,j being an integer greater than zero and less than the number of nodes N n-1 in layer n-1, N n-1 >1, said network further comprising teaching means for setting values for at least one of p n ,j,1, . . . p n ,j,P.sbsb.n,j for each node j in said n'th layer.
Claim 2
Dependent Claim
164 words
A neural network having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each k'th layer having N k nodes, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU13## where i indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,i,j =w.sub.i,j (i, p.sub.n,j,1, . . . ,p.sub.n,j,P.sbsb.n,j), each of said P n ,j being an integer greater than zero and less than the number of nodes N n-1 in layer n-1, N n-1 >1, said network further comprising teaching means for setting values for at least one of p n ,j,1, . . . p n ,j,P.sbsb.n,j for each node j in said n'th layer.
Claim 3
Dependent Claim
24 words
A network according to claim 1, wherein each node in said input layer produces its output value in response to exactly one input value.
Claim 4
Dependent Claim
21 words
A network according to claim 1, wherein all of said layers in said plurality of layers are distinct from each other.
Claim 5
Dependent Claim
43 words
A network according to claim 1, wherein none of the node output values y k1 ,j in a layer k1 of said plurality of layers influences any output value y k0 ,j in a layer k0 of said plurality of layers, where k0<k1.
Claim 6
Dependent Claim
19 words
A network according to claim 1, wherein said plurality of nodes includes all of the nodes in said apparatus.
Claim 7
Dependent Claim
22 words
A network according to claim 1, wherein said plurality of layers collectively includes all of the nodes in said plurality of nodes.
Claim 8
Dependent Claim
13 words
A network according to claim 1, wherein said teaching means comprises back-propagation apparatus.
Claim 9
Dependent Claim
37 words
A network according to claim 1, wherein said teaching means comprises means for setting p n ,j,1, . . . , p n ,j,P.sbsb.n,j to predetermined values in said network for each node j in layer n.
Claim 10
Dependent Claim
25 words
A network according to claim 1, wherein said plurality of layers further includes an output layer different from said input layer and said n'th layer.
Claim 11
Dependent Claim
33 words
A network according to claim 9, wherein f n is a linear function and the node functions for all of said layers except said input layer and said n'th layer are sigmoid functions.
Claim 12
Dependent Claim
13 words
A network according to claim 9, wherein layer n-1 is said input layer.
Claim 13
Dependent Claim
117 words
A network according to claim 1, wherein for each node j in layer n, w n ,j (i, p n ,j,1, . . . ,p n ,j,P.sbsb.n,j)≠0 for at least one value of i between i1 j and i2 j , inclusive, where 1≦i1 j ≦N n-1 , 1≦i2 j ≦N n-1 , and i1 j <i2 j , i1 j and i2 j being such that at least one value of i exists within 1≦i≦N n-1 , but not within i1 j ≦i≦i2 j ; and wherein w n ,j (p n ,j,1, . . . , p n ,j,P.sbsb.n,j)=0 for all values of i within 1≦i≦N n-1 but not within i1 j ≦i≦i2 j .
Claim 14
Dependent Claim
39 words
A network according to claim 12, wherein for each node j in layer n, w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j)≠0for all values of i between i1 j and i2 j inclusive.
Claim 15
Dependent Claim
32 words
A network according to claim 1, wherein for all nodes j in layer n, w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j) is convex in i.
Claim 16
Dependent Claim
31 words
A network according to claim 14, wherein for all nodes j in layer n, w n ,j (i, p n ,j,1, . . . ,p n ,j,P.sbsb.n,j) is bell-shaped in i.
Claim 17
Dependent Claim
20 words
A network according to claim 1, wherein for all nodes j in layer n, P n ,j =3 and ##EQU14##
Claim 18
Dependent Claim
11 words
A network according to claim 16, wherein f n is linear.
Claim 19
Dependent Claim
20 words
A network according to claim 1, wherein for all nodes j in layer n, P n ,j =3 and ##EQU15##
Claim 20
Dependent Claim
18 words
A network according to claim 18, wherein for all nodes j in layer n, p n ,j,3 =2.
Claim 21
Dependent Claim
161 words
A neural network having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each k'th layer having N k nodes, the (n-1)'th layer having more than one node, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU16## indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,i,j =w.sub.n,j (i, p.sub.n,j,1, . . . , p.sub.n,j,P.sbsb.n,j), P n ,j being an integer greater than 0 and w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j) being bell-shaped in i; said network further comprising teaching means for setting values for p n ,j,1, . . . , p n ,j,P.sbsb.n,j.
Claim 22
Dependent Claim
161 words
A neural network having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each k'th layer having N k nodes, the (n-1)'th layer having more than one node, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU16## indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,i,j =w.sub.n,j (i, p.sub.n,j,1, . . . , p.sub.n,j,P.sbsb.n,j), P n ,j being an integer greater than 0 and w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j) being bell-shaped in i; said network further comprising teaching means for setting values for p n ,j,1, . . . , p n ,j,P.sbsb.n,j.
Claim 23
Dependent Claim
51 words
A computational network having a plurality of nodes each having an output carrying a respective output value y, for use with N input signals each having a respective value x i , N>1, each j'th one of said nodes producing its output value y j according to the node function ##EQU17##
Claim 24
Dependent Claim
51 words
A computational network having a plurality of nodes each having an output carrying a respective output value y, for use with N input signals each having a respective value x i , N>1, each j'th one of said nodes producing its output value y j according to the node function ##EQU17##
Claim 25
Dependent Claim
10 words
A network according to claim 21, wherein f is linear.
Claim 26
Dependent Claim
30 words
A network according to claim 21, further comprising teaching means for setting values for a j , b j and c j , for each j'th one of said nodes.
Claim 27
Dependent Claim
51 words
A computational network having a plurality of nodes each having an output carrying a respective output value y, for use with N input signal each having a respective value x i , N>1, each j'th one of said nodes producing its output value y j according to the node function ##EQU18##
Claim 28
Dependent Claim
51 words
A computational network having a plurality of nodes each having an output carrying a respective output value y, for use with N input signal each having a respective value x i , N>1, each j'th one of said nodes producing its output value y j according to the node function ##EQU18##
Claim 29
Independent Claim
181 words
A method for analyzing a plurality of input signals, comprising the steps of: providing adaptive learning apparatus having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each node in said input layer receiving a respective one of said input signals, each k'th layer having N k nodes, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU19## where i indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,i,j =w.sub.n,j (i, p.sub.n,j, i, . . . P.sub.n,j,P.sbsb.n,j), p n ,j being an integer greater than zero and less than the number of nodes N n-1 in layer n-1, N n-1 >1; and training said apparatus by setting values for at least one of p n ,j,1, . . . , p n ,j,P.sbsb.n,j.
Claim 30
Independent Claim
181 words
A method for analyzing a plurality of input signals, comprising the steps of: providing adaptive learning apparatus having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each node in said input layer receiving a respective one of said input signals, each k'th layer having N k nodes, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU19## where i indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,i,j =w.sub.n,j (i, p.sub.n,j, i, . . . P.sub.n,j,P.sbsb.n,j), p n ,j being an integer greater than zero and less than the number of nodes N n-1 in layer n-1, N n-1 >1; and training said apparatus by setting values for at least one of p n ,j,1, . . . , p n ,j,P.sbsb.n,j.
Claim 31
Dependent Claim
43 words
A method according to claim 25, wherein none of the node output values y k1 ,j in a layer k1 of said plurality of layers influences any output value y k0 ,j in a layer k0 of said plurality of layers, where k0<k1.
Claim 32
Dependent Claim
19 words
A method according to claim 25, wherein said training step comprises the step of training said apparatus using back-propagation.
Claim 33
Dependent Claim
38 words
A method according to claim 25, wherein said training step comprises the step setting p n ,j,1, . . . , and p n ,j,P.sbsb.n,j to predetermined values in said apparatus for each node j in layer n.
Claim 34
Dependent Claim
38 words
A method according to claim 28, wherein said training step further comprises the step of predetermining said predetermined values for p n ,j,1, . . . , and p n ,j,P.sbsb.n,j for each node j in layer n.
Claim 35
Dependent Claim
25 words
A method according to claim 25, wherein said plurality of layers further includes an output layer different from said input layer and said n'th layer.
Claim 36
Dependent Claim
33 words
A method according to claim 30, wherein f n is a linear function and the node functions for all of said layers except said input layer and said n'th layer are sigmoid functions.
Claim 37
Dependent Claim
13 words
A method according to claim 31, wherein layer n-1 is said input layer.
Claim 38
Dependent Claim
117 words
A method according to claim 25, wherein for each node j in layer n, w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j)≠0 for all values of i between i1 j and i2 j , inclusive, where 1≦i1 j ≦N n-1 , 1≦i2 j ≦N n-1 , and i1 j <i2 j , i1 j and i2 j being such that at least one value of i exists within 1≦i≦N n-1 , but not within i1 j ≦i≦i2 j ; and wherein w n ,j i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j for all values of i within 1≦i≦N n-1 but not within i1 j ≦i≦i2 j .
Claim 39
Dependent Claim
32 words
A method according to claim 25, wherein for all nodes j in layer n, w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j) is convex in i.
Claim 40
Dependent Claim
32 words
A method according to claim 34, wherein for all nodes j in layer n, w n ,j (i, p n ,j,1, . . . , p n ,j,P.sbsb.n,j) is bell-shaped in i.
Claim 41
Dependent Claim
20 words
A method according to claim 25, wherein for all nodes j in layer n, p n ,j =3 and ##EQU20##
Claim 42
Dependent Claim
20 words
A method according to claim 25, wherein for all nodes j in layer n, p n ,j =3 and ##EQU21##
Claim 43
Independent Claim
188 words
A method for analyzing a plurality of input signals, comprising the steps of: providing adaptive learning apparatus having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each node in said input layer receiving a respective one of said input signals, each k'th layer having N k nodes, the (n-1)'th layer having more than one node, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU22## where i indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,j (i, p.sub.n,j,1, . . . , p.sub.n,j,P.sbsb.n,j) P n ,j being an integer greater than zero and w n ,j (i,p n ,j,1, . . . , p n ,j,P.sbsb.n,j) being bell-shaped in i; and training said apparatus by setting values for p n ,j,1, . . . , and p n ,j,P.sbsb.n,j.
Claim 44
Independent Claim
188 words
A method for analyzing a plurality of input signals, comprising the steps of: providing adaptive learning apparatus having a plurality of nodes each having an output carrying a respective output value y, said nodes being grouped into a plurality of layers including an input layer and an n'th layer different from said input layer, each node in said input layer receiving a respective one of said input signals, each k'th layer having N k nodes, the (n-1)'th layer having more than one node, and each j'th node in each k'th layer except said input layer producing its output value y k ,j according to the node function ##EQU22## where i indexes the nodes of layer k-1 and all the w k ,i,j are connection weights, wherein for all nodes j in said n'th layer, w.sub.n,j (i, p.sub.n,j,1, . . . , p.sub.n,j,P.sbsb.n,j) P n ,j being an integer greater than zero and w n ,j (i,p n ,j,1, . . . , p n ,j,P.sbsb.n,j) being bell-shaped in i; and training said apparatus by setting values for p n ,j,1, . . . , and p n ,j,P.sbsb.n,j.
Claim 45
Independent Claim
84 words
A method for analyzing a plurality of N input signals each having a respective value x i , comprising the steps of: providing a computational network having a plurality of nodes each having an output carrying a respective output value y, each j'th one of said nodes producing its output value Y j according to the node function ##EQU23## and training said computational network by setting values for a j , b j and c j , for each j'th one of said nodes.
Claim 46
Independent Claim
84 words
A method for analyzing a plurality of N input signals each having a respective value x i , comprising the steps of: providing a computational network having a plurality of nodes each having an output carrying a respective output value y, each j'th one of said nodes producing its output value Y j according to the node function ##EQU23## and training said computational network by setting values for a j , b j and c j , for each j'th one of said nodes.
Claim 47
Dependent Claim
10 words
A method according to claim 39, wherein f is linear.
Claim 48
Independent Claim
101 words
A method for analyzing a plurality of input signals each having a respective value x i , comprising the steps of: providing a computational network having a plurality of nodes each having an output carrying a respective output value y, for use with N input signal each having a respective value x i , each j'th one of said nodes producing its output value y j according to the node function ##EQU24## and training said computational network by setting values for at least one of a j , b j and c j , for each j'th one of said nodes.
Claim 49
Independent Claim
101 words
A method for analyzing a plurality of input signals each having a respective value x i , comprising the steps of: providing a computational network having a plurality of nodes each having an output carrying a respective output value y, for use with N input signal each having a respective value x i , each j'th one of said nodes producing its output value y j according to the node function ##EQU24## and training said computational network by setting values for at least one of a j , b j and c j , for each j'th one of said nodes.
Claim 50
Independent Claim
158 words
A method for use with a plurality of N 0 input signals x i produced by a system to be analyzed, comprising the steps of: selecting a parameterized membership function of said input signals; providing adaptive learning apparatus having a plurality of nodes each having an output carrying a respective output value y, each j'th one of said nodes producing its output value y j according to the node function ##EQU25## where i indexes the input signals, wherein for all nodes j, w.sub.i,j =w.sub.j (i, p.sub.j,1, . . . , p.sub.j,p) the parameters p j ,1, . . . , and p j ,p being trainable parameters, P being an integer greater than zero; training said apparatus by setting values for said parameters p j ,1, . . . , p j ,p for each of said nodes j; and observing said trained values for said parameters p j ,1, . . . , and p j ,p.
Claim 51
Independent Claim
158 words
A method for use with a plurality of N 0 input signals x i produced by a system to be analyzed, comprising the steps of: selecting a parameterized membership function of said input signals; providing adaptive learning apparatus having a plurality of nodes each having an output carrying a respective output value y, each j'th one of said nodes producing its output value y j according to the node function ##EQU25## where i indexes the input signals, wherein for all nodes j, w.sub.i,j =w.sub.j (i, p.sub.j,1, . . . , p.sub.j,p) the parameters p j ,1, . . . , and p j ,p being trainable parameters, P being an integer greater than zero; training said apparatus by setting values for said parameters p j ,1, . . . , p j ,p for each of said nodes j; and observing said trained values for said parameters p j ,1, . . . , and p j ,p.
Claim 52
Dependent Claim
9 words
A product produced by the method of claim 25.
Claim 53
Dependent Claim
95 words
A method according to claim 39, further comprising the steps of: producing a plasma adjacent to a subject surface, said plasma producing an optical frequency spectrum dependent upon a plurality of process parameters; forming an amplitude value corresponding to each of a plurality of wavelengths in said optical frequency spectrum; providing each of said amplitude values as respective ones of said input signals x i ; and after said step of training, producing a control signal for one of said process parameters in response to the output value of at least one of said nodes.
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) | 2000-09-01 | 2001-03-03 | 2001-09-02 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2004-09-01 | 2005-03-03 | 2005-09-02 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2008-09-01 | 2009-03-03 | 2009-09-02 | $7700.00 | Window Closed / Lapsed |