Neural network for processing both spatial and temporal data with time based back-propagation
First Claim
1. A processing element (i) for use in a space-time neural network for processing both spacial and temporal data, wherein the neural network comprises a plurality of layers of said processing elements, the plurality of layers comprising a first layer and at least one additional layer, the network further comprising connections between processing elements of the first layer and processing elements of an additional layer:
- each said processing element adapted to receive a sequence of signal inputs X(n), X(n-1), X(n-2) . . . , each input X(n) comprising K signal components x1 (n), x2 (n), . . . xj (n), . . . xk (n), each said processing element comprising, in combination;
(a) a plurality K of adaptable filters (F1i, F2i, . . . Fji, . . . Fki) each filter Fji having an input for receiving a respective component xj (n), xj (n-1), xj (n-2), . . . , of said sequence of inputs, where xj (n) is the most current input component, and providing a filter output yj (n) in response to the input xj (n) which is given by;
space="preserve" listing-type="equation">y.sub.j (n)=f(a.sub.mj Y.sub.j (n-m), b.sub.kj X.sub.j (n-k)),where amj and bkj are coefficients of the filter Fji and f denotes the operation of the filter;
(b) a junction, coupled to each of said adaptive filters, providing a non-linear output pi (Si (n)) in response to the filter outputs yj (n) which is given by;
space="preserve" listing-type="equation">p.sub.i (S.sub.i (n))=f(y.sub.j (n)),where Si (n) is the sum of the filter outputs, whereby said junction presents a sequence of output signals, pi (Si (n)), pi (Si (n-1)), pi (Si (n-2)).
1 Assignment
0 Petitions
Accused Products
Abstract
Neural network algorithms have impressively demonstrated the capability of modelling spatial information. On the other hand, the application of parallel distributed models to processing of temporal data has been severely restricted. The invention introduces a novel technique which adds the dimension of time to the well known back-propagatio
The invention described herein was made by employees of the United States Government and ma be manufactured and used by or for the Government of the United States of America for governmental purposes without payment of any royalties thereon or therefor.
89 Citations
42 Claims
-
1. A processing element (i) for use in a space-time neural network for processing both spacial and temporal data, wherein the neural network comprises a plurality of layers of said processing elements, the plurality of layers comprising a first layer and at least one additional layer, the network further comprising connections between processing elements of the first layer and processing elements of an additional layer:
- each said processing element adapted to receive a sequence of signal inputs X(n), X(n-1), X(n-2) . . . , each input X(n) comprising K signal components x1 (n), x2 (n), . . . xj (n), . . . xk (n), each said processing element comprising, in combination;
(a) a plurality K of adaptable filters (F1i, F2i, . . . Fji, . . . Fki) each filter Fji having an input for receiving a respective component xj (n), xj (n-1), xj (n-2), . . . , of said sequence of inputs, where xj (n) is the most current input component, and providing a filter output yj (n) in response to the input xj (n) which is given by;
space="preserve" listing-type="equation">y.sub.j (n)=f(a.sub.mj Y.sub.j (n-m), b.sub.kj X.sub.j (n-k)),where amj and bkj are coefficients of the filter Fji and f denotes the operation of the filter; (b) a junction, coupled to each of said adaptive filters, providing a non-linear output pi (Si (n)) in response to the filter outputs yj (n) which is given by;
space="preserve" listing-type="equation">p.sub.i (S.sub.i (n))=f(y.sub.j (n)),where Si (n) is the sum of the filter outputs, whereby said junction presents a sequence of output signals, pi (Si (n)), pi (Si (n-1)), pi (Si (n-2)). - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
- each said processing element adapted to receive a sequence of signal inputs X(n), X(n-1), X(n-2) . . . , each input X(n) comprising K signal components x1 (n), x2 (n), . . . xj (n), . . . xk (n), each said processing element comprising, in combination;
-
19. A neural network for processing both spacial and temporal data, wherein said neural network comprises a plurality of layers of processing elements, the plurality of layers comprising a first layer and a second layer, the network further comprising connections between processing elements of the first layer and processing elements of the second layer;
- said first layer of said network adapted to receive a sequence of signal inputs X(n), X(n-1), X(n-2) . . . , each input X(n) comprising N signal components x1 (n), x2 (n), . . . xj (n), . . . xN (n), said first layer of said network comprising, in combination;
(a) a plurality L of first processing elements, each first processing element (i) comprising a plurality N of adaptable filters (F1i, F2i, . . . Fji, . . . FNi), each filter Fji having an input for receiving a respective component xj (n), xj (n-1), xj (n-2), . . . , of said sequence of inputs, where xj (n) is the current input component, and providing a filter output yj (n) in response to an input xj (n) which is given by;
space="preserve" listing-type="equation">y.sub.j (n)=f(a.sub.mj y.sub.j (n-m), b.sub.kj x.sub.j (n-k)),where amj and bkj are coefficients of the filter Fji and f denotes the action of the filter; each first processing element (i) further comprising a first junction, coupled to each of said adaptive filters, providing a non-linear output pi (Si (n)) in response to the filter outputs yj (n) which is given by;
space="preserve" listing-type="equation">p.sub.i (S.sub.i (n))=f(y.sub.j (n)),where Si (n) is the sum of the filter outputs, each first junction presenting a sequence of first output signals, pi (Si (n)), pi (n-1)), pi (Si (n-2)), . . . . - View Dependent Claims (20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42)
- said first layer of said network adapted to receive a sequence of signal inputs X(n), X(n-1), X(n-2) . . . , each input X(n) comprising N signal components x1 (n), x2 (n), . . . xj (n), . . . xN (n), said first layer of said network comprising, in combination;
Specification