US4748674A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Pattern learning and recognition device
Assignee: University of California
•
Filing: 1986-10-07
•
Granted: 1988-05-31
•
Expires: 2006-10-07
Abstract
Abstract A pattern learning and recognition device. The device includes a parallel processing network for receiving and processing an N-dimensional vector input, and a control system for calculating, and storing output signal correlation information during the learning mode of the device, and for distributing the learned information to the network, in the form of feedback gain, during the pattern-recognition mode of the device. The network is preferably composed of an array of N oscillatory units designed to bifurcate from a rest to an active state when stimulated with a threshold input voltage surge.
Inventors
- 👤 Walter J. Freeman
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-24 14:04
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/04; G06N3/049; G06N20/00; G06V; G06V30/00; G06V30/10; G06V30/19; G06V30/192; G06V30/194
Claims Analysis
Total Claims: 14
Claim 1
Independent Claim
156 words
A parallel-circuit signal processing device comprising a series of N integrative units U j , each constructed to respond to the jth component of an N-dimensional input vector, to produce a time varying output signal O j , a multiplexer for sequentially sampling the O j values in each of the N receptor units, means storing a matrix of k ee jk coupling coefficients relating the output signal coupling between pairs of U j , U k units in the device, means for calculating, for each O j value sampled by said multiplexer, feedback input signals which relate the just-sampled O j value and the k ee jk coupling coefficients between U j and other units in the device, and a demultiplexer for applying, following each O j value sampled by the multiplexer, the feedback signals just calculated on the basis of the sampled O j value, to each of the corresponding units U k .
Claim 2
Dependent Claim
156 words
A parallel-circuit signal processing device comprising a series of N integrative units U j , each constructed to respond to the jth component of an N-dimensional input vector, to produce a time varying output signal O j , a multiplexer for sequentially sampling the O j values in each of the N receptor units, means storing a matrix of k ee jk coupling coefficients relating the output signal coupling between pairs of U j , U k units in the device, means for calculating, for each O j value sampled by said multiplexer, feedback input signals which relate the just-sampled O j value and the k ee jk coupling coefficients between U j and other units in the device, and a demultiplexer for applying, following each O j value sampled by the multiplexer, the feedback signals just calculated on the basis of the sampled O j value, to each of the corresponding units U k .
Claim 3
Dependent Claim
73 words
The device of claim 1, for use in learning a class of patterns which can be represented by one or more related N-dimensional vectors, which further includes means for determining coupling coefficients k ee jk between pairs of output signals O j , U k , based on the cross-correlation of the pairs of output signals O j , U k produced in response of the device to one of such related vectors.
Claim 4
Dependent Claim
73 words
The device of claim 1, wherein the integrative unit includes an integrator whose output is connected to a pair of opposed-polarity diodes in the unit, to produce an output signal O j with bilateral saturation characteristics, the resting bias of the unit is substantially below the input voltage level which produces maximal gain in the unit, and each unit, upon receiving its excitatory-level input vector component, is shifted to a different gain condition.
Claim 5
Dependent Claim
47 words
The device of claim 3, wherein the input vector components are surge voltages and those units which are excited from a resting bias to maximal or near-maximal gain levels by input level components return to the resting bias condition after cessation of the inputed component voltage pulses.
Claim 6
Dependent Claim
44 words
The device of claim 4, wherein the integrative units are constructed to produce a monotonic or oscillatory signal upon shift of the gain in the individual units to above-bifurcation levels, and to revert to resting states when the gain reverts to its resting bias.
Claim 7
Dependent Claim
37 words
The device of claim 5, wherein each integrative unit is composed of an excitatory element, and inhibitory element, and a negative feedback connection between the two, and the input vector component is applied to the excitatory element.
Claim 8
Dependent Claim
24 words
The device of claim 6, wherein each excitatory and inhibitory element in an integrator whose signal is passed through a pair of opposed-polarity diodes.
Claim 9
Dependent Claim
37 words
The device of claim 5, wherein the excitatory element is coupled in negative feedback with itself or a second excitatory unit, and the inhibitory element is coupled in negative feedback with itself or a second inhibitory element.
Claim 10
Dependent Claim
110 words
The device of claim 5, for use in associating an input pattern with a learned stereotypic pattern, where the N-dimensional vector representing the learned pattern produces a limit cycle attractor in the device which is characterized by a given pattern of oscillation in the N units, wherein the N-dimensional pattern representing the input pattern produces a bifurcation in the device to the limit attractor cycle, in which case the new pattern is recognized as related to the learned pattern, or to some other attractor cycle characterized by a different pattern of output oscillations in the units, in which case the pattern is recognized as not related to the learned pattern.
Claim 11
Dependent Claim
18 words
The device of claim 1, wherein the rest state of the device is a non-oscillatory point attractor state.
Claim 12
Dependent Claim
25 words
The device of claim 1, wherein the rest state of the device is a chaotic cycle attractor characterized by random oscillatory signals in the units.
Claim 13
Independent Claim
148 words
The method of processing patterns which are representable as N-dimensional vectors comprising providing a series fo N receptor units U j , each constructed to respond to the ith component of an N-dimensional input vector, to produce a time varying output signal O j , sequentially sampling the O j values in each of the N receptor units, storing a matrix of k ee jk coupling coefficients relating the output signal coupling between pairs of U j , U k units in the device, calculating, for each O j value sampled, feedback input signals which relate the just-sample O j value and the k ee jk coupling coefficients between U i and other units in the device, and following each O j value sampled, applying the feedback signals just calculated on the basis of the sampled O j value, to each of the corresponding units U k .
Claim 14
Independent Claim
148 words
The method of processing patterns which are representable as N-dimensional vectors comprising providing a series fo N receptor units U j , each constructed to respond to the ith component of an N-dimensional input vector, to produce a time varying output signal O j , sequentially sampling the O j values in each of the N receptor units, storing a matrix of k ee jk coupling coefficients relating the output signal coupling between pairs of U j , U k units in the device, calculating, for each O j value sampled, feedback input signals which relate the just-sample O j value and the k ee jk coupling coefficients between U i and other units in the device, and following each O j value sampled, applying the feedback signals just calculated on the basis of the sampled O j value, to each of the corresponding units U k .
References Cited (Backward Citations)
US3646329A
Cited
US4504970A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1991-05-31 | 1991-11-30 | 1992-05-31 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1995-05-31 | 1995-11-30 | 1996-05-31 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 1999-05-31 | 1999-11-30 | 2000-05-31 | $7700.00 | Window Closed / Lapsed |