US5103496A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Artificial neural network system for memory modification
Assignee: United States Department of the Navy
•
Filing: 1991-06-27
•
Granted: 1992-04-07
•
Expires: 2011-06-27
Abstract
Abstract An artificial neural network, which has a plurality of neurons each receiving a plurality of inputs whose effect is determined by adjust able weights at synapses individually connecting the inputs to the neuron to provide a sum signal to a sigmoidal function generator determining the output of the neuron, undergoes memory modification by a steepest-descent method in which individual variations in the outputs of the neurons are successively generated by small perturbations imposed on the sum signals. As each variation is generated on the output of a neuron, an overall error of all the neuron outputs in relation to their desired values is measured and compared to this error prior to the perturbation. The difference in these errors, with adjustments which may be changed as the neuron outputs converge toward their desired values, is used to modify each weight of the neuron presently subjected to the perturbation.
Inventors
- 👤 David K. Andes
- 👤 Robert A. Licklider
- 👤 Donald H. Witcher
- 👤 Richard M. Swenson
- 👤 James F. Barbieri
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-06 08:28
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063
Claims Analysis
Total Claims: 6
Claim 1
Independent Claim
224 words
An artificial neural network system adapted for memory modification and comprising: a plurality of inputs; a plurality of summing circuits; a plurality of variable conductances connecting said inputs individually to said summing circuits; means for generating a plurality of output signals individual to said summing circuits, each output signal being determined from a signal provided by the corresponding summing circuit in accordance with a monotonically non decreasing and differentiable function providing upper and lower bounds for the output signal; and means for receiving a plurality of desired signals individual to said summing circuits and for generating an error signal representing the differences between every output signal and the desired signal corresponding thereto; means for successively addressing said summing circuits and for providing a perturbation to an addressed summing circuit; means for generating a correction which is inversely proportional to said perturbation, which is directly proportional to a change in said error signal caused by provision of said perturbation to an addressed summing circuit, and which is proportional to a gain factor; and means for addressing the variable conductances of said plurality thereof corresponding to an addressed summing circuit and for adjusting each addressed conductance by a variation which is proportional to said correction and which corresponds to an activation signal provided by the one of said inputs corresponding to each of the addressed conductances.
Claim 2
Dependent Claim
224 words
An artificial neural network system adapted for memory modification and comprising: a plurality of inputs; a plurality of summing circuits; a plurality of variable conductances connecting said inputs individually to said summing circuits; means for generating a plurality of output signals individual to said summing circuits, each output signal being determined from a signal provided by the corresponding summing circuit in accordance with a monotonically non decreasing and differentiable function providing upper and lower bounds for the output signal; and means for receiving a plurality of desired signals individual to said summing circuits and for generating an error signal representing the differences between every output signal and the desired signal corresponding thereto; means for successively addressing said summing circuits and for providing a perturbation to an addressed summing circuit; means for generating a correction which is inversely proportional to said perturbation, which is directly proportional to a change in said error signal caused by provision of said perturbation to an addressed summing circuit, and which is proportional to a gain factor; and means for addressing the variable conductances of said plurality thereof corresponding to an addressed summing circuit and for adjusting each addressed conductance by a variation which is proportional to said correction and which corresponds to an activation signal provided by the one of said inputs corresponding to each of the addressed conductances.
Claim 3
Dependent Claim
135 words
The artificial neural network system of claim 1 wherein said output signals, said desired signals, and said activation signal have corresponding signed magnitudes and wherein: said generating of said error signal includes summing a plurality of signals each determined by the absolute magnitude of one of said differences; said generating of said correction includes determining said change in said error signal as a signed magnitude having a sign which is defined by the direction of said change when such a perturbation is provided to an addressed summing circuit; and said adjusting of an addressed conductance includes determining said variation as a signed magnitude having a sign which defines the direction of said adjusting of an addressed conductance and which is determined by the sign of said change and by the sign of said activation signal.
Claim 4
Dependent Claim
33 words
The artificial neural network system of claim 1 wherein said variation is determined as independent of the magnitude of said activation signal and is determined as proportional to the magnitude of said correction.
Claim 5
Dependent Claim
34 words
The artificial neural network system of claim 1 wherein the magnitude of said variation is determined as proportional to the magnitude of said activation signal and as proportional to the magnitude of said correction.
Claim 6
Dependent Claim
23 words
The artificial neural network system of claim 1 wherein the magnitude of said perturbation is proportional to the magnitude of said error signal.
References Cited (Backward Citations)
US4660166A
Cited
US4719591A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1995-04-07 | 1995-10-07 | 1996-04-07 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1999-04-07 | 1999-10-07 | 2000-04-07 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2003-04-07 | 2003-10-07 | 2004-04-07 | $7700.00 | Window Closed / Lapsed |