Intelligent system for automatic feature detection and selection or identification
First Claim
1. 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 Nk nodes, and each j'"'"'th node in each k'"'"'th layer except said input layer producing its output value yk,j according to the node function ##EQU13## where i indexes the nodes of layer k-1 and all the wk,i,j are connection weights,wherein for all nodes j in said n'"'"'th layer,
each of said Pn,j being an integer greater than zero and less than the number of nodes Nn-1 in layer n-1, Nn-1 >
1,said network further comprising teaching means for setting values for at least one of pn,j,1, . . . pn,j,P.sbsb.n,j for each node j in said n'"'"'th layer.
2 Assignments
0 Petitions
Accused Products
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 ##EQU1## where Nk-1 is the number of nodes in layer 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,p.sbsb.n).
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 ##EQU2##
71 Citations
44 Claims
- 1. 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 Nk nodes, and each j'"'"'th node in each k'"'"'th layer except said input layer producing its output value yk,j according to the node function ##EQU13## where i indexes the nodes of layer k-1 and all the wk,i,j are connection weights,
wherein for all nodes j in said n'"'"'th layer, - space="preserve" listing-type="equation">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 Pn,j being an integer greater than zero and less than the number of nodes Nn-1 in layer n-1, Nn-1 >
1,said network further comprising teaching means for setting values for at least one of pn,j,1, . . . pn,j,P.sbsb.n,j for each node j in said n'"'"'th layer. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
- 20. 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 Nk 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 yk,j according to the node function ##EQU16## indexes the nodes of layer k-1 and all the wk,i,j are connection weights,
wherein for all nodes j in said n'"'"'th layer, - space="preserve" listing-type="equation">w.sub.n,i,j =w.sub.n,j (i, p.sub.n,j,1, . . . , p.sub.n,j,P.sbsb.n,j),
Pn,j being an integer greater than 0 and wn,j (i, pn,j,1, . . . , pn,j,P.sbsb.n,j) being bell-shaped in i; said network further comprising teaching means for setting values for pn,j,1, . . . , pn,j,P.sbsb.n,j.
-
21. 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 xi, N>
- 1, each j'"'"'th one of said nodes producing its output value yj according to the node function ##EQU17##
- View Dependent Claims (22, 23)
-
24. 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 xi, N>
- 1, each j'"'"'th one of said nodes producing its output value yj according to the node function ##EQU18##
-
25. 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 Nk nodes, and each j'"'"'th node in each k'"'"'th layer except said input layer producing its output value yk,j according to the node function ##EQU19## where i indexes the nodes of layer k-1 and all the wk,i,j are connection weights, wherein for all nodes j in said n'"'"'th layer,
space="preserve" listing-type="equation">w.sub.n,i,j =w.sub.n,j (i, p.sub.n,j, i, . . . P.sub.n,j,P.sbsb.n,j),pn,j being an integer greater than zero and less than the number of nodes Nn-1 in layer n-1, Nn-1 >
1; andtraining said apparatus by setting values for at least one of pn,j,1, . . . , pn,j,P.sbsb.n,j. - View Dependent Claims (26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 43)
-
-
38. 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 Nk 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 yk,j according to the node function ##EQU22## where i indexes the nodes of layer k-1 and all the wk,i,j are connection weights, wherein for all nodes j in said n'"'"'th layer,
space="preserve" listing-type="equation">w.sub.n,j (i, p.sub.n,j,1, . . . , p.sub.n,j,P.sbsb.n,j)Pn,j being an integer greater than zero and wn,j (i,pn,j,1, . . . , pn,j,P.sbsb.n,j) being bell-shaped in i; and training said apparatus by setting values for pn,j,1, . . . , and pn,j,P.sbsb.n,j.
-
-
39. A method for analyzing a plurality of N input signals each having a respective value xi, 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 Yj according to the node function ##EQU23## and training said computational network by setting values for aj, bj and cj, for each j'"'"'th one of said nodes. - View Dependent Claims (40, 44)
-
41. A method for analyzing a plurality of input signals each having a respective value xi, 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 xi, each j'"'"'th one of said nodes producing its output value yj according to the node function ##EQU24## and training said computational network by setting values for at least one of aj, bj and cj, for each j'"'"'th one of said nodes.
-
42. A method for use with a plurality of N0 input signals xi 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 yj according to the node function ##EQU25## where i indexes the input signals, wherein for all nodes j,
space="preserve" listing-type="equation">w.sub.i,j =w.sub.j (i, p.sub.j,1, . . . , p.sub.j,p)the parameters pj,1, . . . , and pj,p being trainable parameters, P being an integer greater than zero; training said apparatus by setting values for said parameters pj,1, . . . , pj,p for each of said nodes j; and observing said trained values for said parameters pj,1, . . . , and pj,p.
-
Specification