US4906865A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Sample and hold circuit for temporal associations in a neural network
Assignee: Intel Corp
•
Filing: 1988-12-09
•
Granted: 1990-03-06
•
Expires: 2008-12-09
Abstract
Abstract A sample and hold circuit for introducing delayed feedback into an associative memory is described. The circuit continuously samples an output sequence derived from a neural network; then, in response to a clock signal, it holds that output sequence until the next clock signal. The held sequence is coupled back to the input of the network so that the present output sequence becomes some function of the past output sequence. This delayed feedback enables the associative recall of a memorized sequence from the neural network.
Inventors
- 👤 Mark A. Holler
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-28 07:59
CPC Classifications
G; G11; G11C; G11C27/00; G11C27/02; G11C27/024; G11C27/026; G11C15/00; G11C15/04
Claims Analysis
Total Claims: 16
Claim 1
Independent Claim
143 words
An associative memory for memorizing and reconstructing time sequences of representations comprising: a neural network having a plurality of synapse inputs and an output providing an output pattern which varies in time, said neural network receiving a continuous sequence of data at said synapse inputs; a nonintegrating sample and hold device having a data input coupled to said output of said network for sampling and holding said output pattern for a certain time period so as to produce a delayed output pattern at a feedback output coupled to at least one of said plurality of synapse inputs to provide feedback of said output pattern to said neural network; and a means for synchronizing said feedback output of said sample and hold device with the time domain of said continuous sequence of data to allow recognition of said continuous sequence by said neural network.
Claim 2
Dependent Claim
143 words
An associative memory for memorizing and reconstructing time sequences of representations comprising: a neural network having a plurality of synapse inputs and an output providing an output pattern which varies in time, said neural network receiving a continuous sequence of data at said synapse inputs; a nonintegrating sample and hold device having a data input coupled to said output of said network for sampling and holding said output pattern for a certain time period so as to produce a delayed output pattern at a feedback output coupled to at least one of said plurality of synapse inputs to provide feedback of said output pattern to said neural network; and a means for synchronizing said feedback output of said sample and hold device with the time domain of said continuous sequence of data to allow recognition of said continuous sequence by said neural network.
Claim 3
Dependent Claim
59 words
The associative memory of claim 1 wherein said plurality of synapse inputs includes a temporal pattern input and a context input such that when said temporal pattern and said context are provided in association with said delayed output pattern, a next output pattern is evoked at said output of said network thereby triggering temporal recall of said time sequence.
Claim 4
Dependent Claim
30 words
The associative memory of claim 2 further comprising a clocking means for providing clock pulses to said device such that the period of said clock pulses determines said time period.
Claim 5
Dependent Claim
185 words
An associative memory for memorizing and reconstructing time sequences of representations comprising: a neural network having a plurality of synapse inputs for continuously receiving input data in the form of a temporal pattern, a context, and a delayed feedback of an output pattern, said neural network also having a plurality of neuron amplifiers and an output providing said output pattern which varies in time; a nonintegrating sample and hold device having a clock input, and a data input coupled to said output of said neural network for sampling and holding said output pattern for a certain time period so as to produce a delayed output pattern at a feedback output coupled to at least one of said plurality of synapse inputs to provide feedback of said output pattern to said neural network; and a clocking means coupled to said clock input for producing clock pulses to said device, said clock pulses providing said delayed feedback of said output pattern such that when said temporal pattern, said context, and said delayed feedback are synchronously, presented to said network, temporal recollection of said time sequence is evoked.
Claim 6
Dependent Claim
185 words
An associative memory for memorizing and reconstructing time sequences of representations comprising: a neural network having a plurality of synapse inputs for continuously receiving input data in the form of a temporal pattern, a context, and a delayed feedback of an output pattern, said neural network also having a plurality of neuron amplifiers and an output providing said output pattern which varies in time; a nonintegrating sample and hold device having a clock input, and a data input coupled to said output of said neural network for sampling and holding said output pattern for a certain time period so as to produce a delayed output pattern at a feedback output coupled to at least one of said plurality of synapse inputs to provide feedback of said output pattern to said neural network; and a clocking means coupled to said clock input for producing clock pulses to said device, said clock pulses providing said delayed feedback of said output pattern such that when said temporal pattern, said context, and said delayed feedback are synchronously, presented to said network, temporal recollection of said time sequence is evoked.
Claim 7
Dependent Claim
20 words
The associative memory of claim 4 wherein the period between said clock pulses determines the delay in said delayed feedback.
Claim 8
Dependent Claim
27 words
The associative memory of claim 5 further comprising a reset means for placing said at least one of said plurality of synapse inputs at a predetermined potential.
Claim 9
Dependent Claim
316 words
In a neural network having a plurality of neural amplifiers and a plurality of synapse inputs for receiving a continuous sequence of data in the form of a temporal pattern and a context, a nonintegrating sample and hold circuit providing delayed feedback of an output pattern from said network comprising: a first storage member having a clock input, a data input, and an output, said data input coupled to said output pattern of said network; a differential amplifier having a first input coupled to said output of said first storage member and a second input coupled to a reference potential, said amplifier also having a differential output providing single-to-double-ended conversion of said output pattern; a second storage member having a differential input coupled to said differential output of said amplifier, a clock input, and a differential word line coupled to at least one of said plurality of synapse inputs; a clocking means for providing a first clock pulse to said clock input of said first storage member and a second clock pulse to said clock input of said second storage member; said first storage member being responsive to said first pulse such that when said first pulse is absent said first storage member samples a present output pattern and when said first pulse is applied said first storage member holds said present output pattern, said second storage member being responsive to said second pulse such that when said second pulse is absent said second storage member holds a past one of said output pattern and when said second pulse is active said second storage member samples said present output pattern, amplified by said amplifier such that said word line output represents a delayed feedback of said output pattern, said first and second pulses being applied synchronous with said temporal pattern and said context inputs so as to facilitate temporal recall of a memorized sequence of representations.
Claim 10
Dependent Claim
316 words
In a neural network having a plurality of neural amplifiers and a plurality of synapse inputs for receiving a continuous sequence of data in the form of a temporal pattern and a context, a nonintegrating sample and hold circuit providing delayed feedback of an output pattern from said network comprising: a first storage member having a clock input, a data input, and an output, said data input coupled to said output pattern of said network; a differential amplifier having a first input coupled to said output of said first storage member and a second input coupled to a reference potential, said amplifier also having a differential output providing single-to-double-ended conversion of said output pattern; a second storage member having a differential input coupled to said differential output of said amplifier, a clock input, and a differential word line coupled to at least one of said plurality of synapse inputs; a clocking means for providing a first clock pulse to said clock input of said first storage member and a second clock pulse to said clock input of said second storage member; said first storage member being responsive to said first pulse such that when said first pulse is absent said first storage member samples a present output pattern and when said first pulse is applied said first storage member holds said present output pattern, said second storage member being responsive to said second pulse such that when said second pulse is absent said second storage member holds a past one of said output pattern and when said second pulse is active said second storage member samples said present output pattern, amplified by said amplifier such that said word line output represents a delayed feedback of said output pattern, said first and second pulses being applied synchronous with said temporal pattern and said context inputs so as to facilitate temporal recall of a memorized sequence of representations.
Claim 11
Dependent Claim
44 words
The sample and hold circuit of claim 7, wherein said first storage member comprises a transmission gate coupled in parallel to a capacitor in said second storage member comprises a pair of transmission gates each of which is coupled in parallel to a capacitor.
Claim 12
Dependent Claim
29 words
The sample and hold circuit of claim 7, further comprising a reset means coupled to said second storage member for placing a predetermined potential on said word line output.
Claim 13
Independent Claim
99 words
A method of providing delayed feedback of an output pattern in a neural network to facilitate a temporal recall of a memorized sequence of representations, said neural network receiving a continuous sequence of data in the form of a temporal pattern and a context said method comprising the steps of: sampling said output pattern from said neural network; holding said output pattern for a certain time period thereby producing a delayed output pattern; inputting said delayed output pattern into said neural network synchronous with said temporal pattern and said context to evoke the next sequence in said memorized sequence.
Claim 14
Independent Claim
99 words
A method of providing delayed feedback of an output pattern in a neural network to facilitate a temporal recall of a memorized sequence of representations, said neural network receiving a continuous sequence of data in the form of a temporal pattern and a context said method comprising the steps of: sampling said output pattern from said neural network; holding said output pattern for a certain time period thereby producing a delayed output pattern; inputting said delayed output pattern into said neural network synchronous with said temporal pattern and said context to evoke the next sequence in said memorized sequence.
Claim 15
Independent Claim
140 words
A method of providing delayed feedback of an output pattern in a neural network to facilitate temporal recall of a memorized sequence of representations, said neural network receiving a continuous sequence of data in the form of a temporal pattern and a context input into a plurality of synapses, said method comprising the steps of: sampling said output pattern from said neural network; holding said output pattern in response to a first clock pulse; differentially amplifying said output pattern being held; storing said amplified output pattern in response to a second clock pulse for a certain time period, thereby producing a delayed output pattern; inputting said delayed output pattern into at least one of said plurality of synapse inputs in said neural network synchronous with said temporal pattern and said context to evoke a next pattern in said memorized sequence.
Claim 16
Independent Claim
140 words
A method of providing delayed feedback of an output pattern in a neural network to facilitate temporal recall of a memorized sequence of representations, said neural network receiving a continuous sequence of data in the form of a temporal pattern and a context input into a plurality of synapses, said method comprising the steps of: sampling said output pattern from said neural network; holding said output pattern in response to a first clock pulse; differentially amplifying said output pattern being held; storing said amplified output pattern in response to a second clock pulse for a certain time period, thereby producing a delayed output pattern; inputting said delayed output pattern into at least one of said plurality of synapse inputs in said neural network synchronous with said temporal pattern and said context to evoke a next pattern in said memorized sequence.
References Cited (Backward Citations)
US3103648A
Cited
US3950733A
Cited
US4660166A
Cited
US4760437A
Cited
US4773024A
Cited
US4802103A
Cited
US4807168A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1993-03-05 | 1993-09-04 | 1994-03-06 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1997-03-05 | 1997-09-04 | 1998-03-06 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2001-03-05 | 2001-09-04 | 2002-03-06 | $7700.00 | Window Closed / Lapsed |