Patent US4540937A successfully refreshed from Google Patents.
Patent EP0242110A2 successfully refreshed from Google Patents.
EP0242110A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Neural network for minimizing an energy function
Assignee: AT&T Inc
•
Filing: 1987-04-07
•
Granted: —
•
Expires: 2007-04-07
Abstract
Abstract Advantageous neural network realizations are achieved by employing only negative gain amplifiers (100) and a clipped T matrix (20) having conductances T ij which have only two values. Preferably, one of these values is a preselected value set by the value of a fixed resistor, and the other value is zero, created simply with an open circuit. Values for the T ij terms of the clipped T matrix are obtained through an iterative process which operates on the clipped and nonclipped matrices and minimizes the error resulting from the use of the clipped T matrix.
Inventors
- 👤 John Stewart Denker
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-10-31 15:46
CPC Classifications
G; G06; G06G; G06G7/00; G06G7/12; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063; G06N3/065
Claims Analysis
Total Claims: 10
Claim 1
Independent Claim
66 words
A neural network comprising: a plurality of amplifiers (10) having an input and an output; means for applying input currents (41) to said input of each of said amplifiers; CHARACTERIZED BY an interconnection matrix (20), connecting said output of each of said amplifiers to each of said inputs with a conductance (T2i) that is either of a first preselected value or of a second preselected value.
Claim 2
Dependent Claim
66 words
A neural network comprising: a plurality of amplifiers (10) having an input and an output; means for applying input currents (41) to said input of each of said amplifiers; CHARACTERIZED BY an interconnection matrix (20), connecting said output of each of said amplifiers to each of said inputs with a conductance (T2i) that is either of a first preselected value or of a second preselected value.
Claim 3
Dependent Claim
14 words
The network of claim 1 wherein one of said preselected values is essentially zero.
Claim 4
Dependent Claim
11 words
The network of claim 1 wherein said amplifiers are inverting amplifiers.
Claim 5
Dependent Claim
119 words
A neural network for minimizing an energy function having a plurality of local minima comprising: a plurality of amplifiers having an input and an inverting output; means for applying input currents to the input of each of said amplifiers; CHARACTERIZED BY an interconnection matrix, connecting the non-inverting output of each of said amplifiers to each said input with a conductance T ij that is either of a first preselected value or of a second preselected value, where said T' ij represents the conductance between the output of the ith amplifier and the input of the jth amplifier and where said T' ij conductances are selected to create said plurality of local minima at which said network is at equilibrium.
Claim 6
Dependent Claim
119 words
A neural network for minimizing an energy function having a plurality of local minima comprising: a plurality of amplifiers having an input and an inverting output; means for applying input currents to the input of each of said amplifiers; CHARACTERIZED BY an interconnection matrix, connecting the non-inverting output of each of said amplifiers to each said input with a conductance T ij that is either of a first preselected value or of a second preselected value, where said T' ij represents the conductance between the output of the ith amplifier and the input of the jth amplifier and where said T' ij conductances are selected to create said plurality of local minima at which said network is at equilibrium.
Claim 7
Dependent Claim
94 words
The network of claim 4 wherein said conductances are selected in accordance with a procedure of developing a set of multi-level T ij values that create an energy function having said plurality of local minima, and of repetitively computing T' ij values which are truncated representations of the selected expression where r is less than one, V equals T'' ij is tne T' ij from tie previous repetition, and V A j is the jth component of one of said local minima said computing of T' ij being done over all of said minima.
Claim 8
Dependent Claim
19 words
The network of claim 5 wherein said truncated representation is realized by recognizing the sign of said selected expression.
Claim 9
Dependent Claim
89 words
The network of claim 4 wherein said conductances are selected in accordance with a procedure of developing a set of multi-level T ij values that create an energy function having said plurality of local minima, and of repetitively computing T 'ij values which are truncated representations of the selected expression where r is less than one, V equals 'ij is the T'ij from the previous repetition, and is the jth component of one of said local minima, said computing of T' ij being done over all of said minima.
Claim 10
Dependent Claim
19 words
The network of claim 7 wherein said truncated representation is realized by recognizing the sign of said selected expression.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.