Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Examiner’s Note
Providing supporting paragraph(s) for each limitation of amended/new claim(s) in Remarks is strongly requested for clear and definite claim interpretations by Examiner (e.g., to avoid rejections under 35 U.S.C § 112(a) “Lack of written description”)
Applicant can schedule interviews (via Automated Interview Request (AIR)) at any stage of the prosecution (e.g., Non-Final, Final, and After-Final) to discuss any issues related to, for example, rejections under 35 U.S.C § 101 and § 102/103, for moving toward allowance.
If a limitation has bold brackets (i.e. [·]) around claim languages, the bracketed claim languages indicate that they have not been taught yet by the current prior art reference but they will be taught by another prior art reference afterwards.
If a limitation has one or more bold underlines, the one or more bold underlined claim languages indicate that they are taught by the current prior art reference, while the one or more non-underlined claim languages indicate that they have been taught already by one or more previous art references.
Priority
Acknowledgment is made of applicant's claim for the present application filed on 10/18/2023.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier.
Such claim limitation(s) is/are:
Claim 19: “the spiking unit receives a first input from the first unit and receives a second input from the second unit” (Note that pars 111-123 of the present application describe a sufficient structure for performing the claimed function.)
Claim 19: “the spiking unit updates a state of the first neuron based on the first input and the second input.” (Note that pars 111-123 of the present application describe a sufficient structure for performing the claimed function.)
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1
Step 1: “Is the claim to a process, machine, manufacture, or composition of matter?”
The claim is directed to a method. Therefore, yes.
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?”
encoding, …, one or more variables of an optimization problem, (i.e., mental process)
modifying the one or more values of the one or more variables by changing the one or more states of the one or more first neurons; (i.e., mental process)
computing, …, a cost using a cost function based on the one or more values modified of the one or more variables; (i.e., mathematical concept)
determining, …, whether the cost meets a convergence criterion; and (i.e., mental process)
The claim is directed to an abstract idea. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
A computer-implemented (well-understood, routine, and conventional generic computer and/or model, see MPEP 2106.05(f))
by one or more first neurons in a neural network (well-understood, routine, and conventional generic computer and/or model, see MPEP 2106.05(f))
one or more states of the one or more first neurons representing one or more values of the one or more variables; (a particular type or source of model/data, Field of Use and Technological Environment, see MPEP 2106.05(h))
transmitting, by the one or more first neurons, one or more spikes to a second neuron in the neural network (insignificant extra-solution activity of transmitting data, see MPEP 2106.05(g)), the one or more spikes comprising one or more modified values of the one or more variables; (a particular type or source of model/data, Field of Use and Technological Environment, see MPEP 2106.05(h))
by the second neuron (well-understood, routine, and conventional generic computer and/or model, see MPEP 2106.05(f))
by a third neuron in the neural network (well-understood, routine, and conventional generic computer and/or model, see MPEP 2106.05(f))
in response to determining that the cost meets the convergence criterion, transmitting, by the third neuron, a message to the one or more first neurons, the message instructing the one or more first neurons to stop changing the one or more states. (insignificant extra-solution activity of transmitting data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Specifically, the claimed inventions simply append well-understood, routine and conventional activities previously known to the industry, both when viewed independently and as an ordered combination, specified at a high level of generality, to the judicial exception, (e.g., a claim to an abstract idea requiring no more than a generic computer to perform generic computer functions that are well-understood, routine and conventional activities previously known to the industry). Therefore, no.
Regarding claim 2
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?” The claim recites the abstract idea identified above regarding claim 1. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
a first neuron comprises a spiking unit, a first unit, and a second unit, (a particular type or source of model/data, Field of Use and Technological Environment, see MPEP 2106.05(h))
the spiking unit receives a first input from the first unit and receives a second input from the second unit, and (insignificant extra-solution activity of receiving data, see MPEP 2106.05(g))
the spiking unit updates a state of the first neuron based on the first input and the second input. (insignificant extra-solution activity of storing data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 3
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?”
the first unit computes the first input based on data received from another first neuron, and (i.e., mathematical concept)
The claim is directed to an abstract idea. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
the second input from the second neuron is a prior state of the first neuron. (a particular type or source of model/data, Field of Use and Technological Environment, see MPEP 2106.05(h))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 4
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?” The claim recites the abstract idea identified above regarding claim 1. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
the first neuron further comprises an additional unit, and (a particular type or source of model/data, Field of Use and Technological Environment, see MPEP 2106.05(h))
the additional unit, based on a message from the third neuron, resets the state of the first neuron to an initialized state of the first neuron (insignificant extra-solution activity of storing data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 5
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?”
the additional unit computes a time-weighted average of states of the first neuron, and (i.e., mathematical concept)
The claim is directed to an abstract idea. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
the first neuron further comprises an additional unit, (a particular type or source of model/data, Field of Use and Technological Environment, see MPEP 2106.05(h))
the spiking unit sends out a spike encoding the state of the first neuron based on a determination that the state of the first neuron is equal to or greater than the time-weighted average (insignificant extra-solution activity of transmitting data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 6
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?” The claim recites the abstract idea identified above regarding claim 1. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
wherein the message further instructs the one or more first neurons to send a processing unit the one or more spikes as a solution to the optimization problem. (insignificant extra-solution activity of transmitting data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 7
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?” The claim recites the abstract idea identified above regarding claim 1. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
in response to determining that the cost fails to meet the convergence criterion, transmitting, by the third neuron, a different message to one or more units in the one or more first neurons (insignificant extra-solution activity of transmitting data, see MPEP 2106.05(g)), the different message instructing the one or more units in the one or more first neurons to further modify the one or more values of the one or more variables by further changing the one or more states of the one or more first neurons. (insignificant extra-solution activity of storing data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 8
Step 1: “Is the claim to a process, machine, manufacture, or composition of matter?”
The claim is directed to a composition of matter. Therefore, yes.
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?”
determining, …, whether a stalling period threshold is reached based on a spike from the third neuron; and (i.e., mental process)
The claim is directed to an abstract idea. Therefore, yes.
Step 2A Prong 2: “Does the claim recite additional elements that integrate the judicial exception into a practical application?” The following elements are directed to additional elements:
by a fourth neuron in the neural network (well-understood, routine, and conventional generic computer and/or model, see MPEP 2106.05(f))
after determining that stalling period threshold is reached, instructing, by the fourth neuron, the third neuron to transmit a different message to the one or more first neurons (insignificant extra-solution activity of transmitting data, see MPEP 2106.05(g)), the different message instructing the one or more first neurons to change the one or more modified values of the one or more variables back to the one or more values of the one or more variables (insignificant extra-solution activity of storing data, see MPEP 2106.05(g))
Therefore, no.
Step 2B: “Does the claim recite additional elements that amount to significantly more than the judicial exception?”
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Therefore, no.
Regarding claim 9
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?”
determining whether the cost is equal to or lower than a target cost (i.e., mental process)
The claim is directed to an abstract idea. Therefore, yes.
The claim does not add any additional elements (Step 2A Prong 2) or significantly more (Step 2B).
Regarding claim 10
Step 2A Prong 1: “Does the claim recite an abstract idea, law of nature, or natural phenomenon?”
determining whether a number of steps in which the one or more first neurons change the one or more states exceeds a threshold number. (i.e., mental process)
The claim is directed to an abstract idea. Therefore, yes.
The claim does not add any additional elements (Step 2A Prong 2) or significantly more (Step 2B).
Regarding claim 11
The claim is rejected for the reasons set forth in the rejection of Claim 1 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 12
The claim is rejected for the reasons set forth in the rejection of Claim 2 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 13
The claim is rejected for the reasons set forth in the rejection of Claim 3 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 14
The claim is rejected for the reasons set forth in the rejection of Claim 4 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 15
The claim is rejected for the reasons set forth in the rejection of Claim 5 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 16
The claim is rejected for the reasons set forth in the rejection of Claim 6 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 17
The claim is rejected for the reasons set forth in the rejection of Claim 7 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 18
The claim is rejected for the reasons set forth in the rejection of Claim 8 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 19
The claim is rejected for the reasons set forth in the rejection of Claim 1 under 35 U.S.C. 101, mutatis mutandis.
Regarding claim 20
The claim is rejected for the reasons set forth in the rejection of Claim 2 under 35 U.S.C. 101, mutatis mutandis.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim(s) 1-4, 6-7, 9-14, 16-17, 19-20 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Alom et al. (Quadratic Unconstrained Binary Optimization (QUBO) on Neuromorphic Computing System).
Regarding claim 1
Alom teaches
A computer-implemented method, comprising:
(Alom [sec(s) Abs] “In this paper, we implemented the NP-hard optimization problem called Quadratic Unconstrained Binary Optimization (QUBO) problem for the solution of graph problems on the IBM’s Neurosynaptic TrueNorth System.” [sec(s) V] “The entire model is written in MATLAB, using the integrated programming environment called corelet programming for IBM’s Neurosynaptic system [16].”;)
encoding, by one or more first neurons in a neural network, one or more variables of an optimization problem, one or more states of the one or more first neurons representing one or more values of the one or more variables;
(Alom [sec(s) II] “According to the definition of the QUBO problem in equations (1) and (2), it has only input values X ϵ {0, 1}n and produces outputs Y ϵ {0, 1}n.” [sec(s) III.B] “The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step.” [sec(s) IV.B] “Four types of corelets are used for different operations: random spike generator, adder corelet, QUBO corelet, and output corelet. The random spike-generator generates the random spikes as inputs to the system. The number of spikes have been generated with respect to the dimensions of the input weight matrix based on number of nodes in an input graph. For example, if the input dimension of weight matrix is 4x4, it means that the number of nodes in the graph is 4. Nodes are connected with each other through positive or negative weights as shown in Fig 3. Thus the number of input spikes is 4 for this graph.”;)
modifying the one or more values of the one or more variables by changing the one or more states of the one or more first neurons;
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.”;)
transmitting, by the one or more first neurons, one or more spikes to a second neuron in the neural network, the one or more spikes comprising one or more modified values of the one or more variables;
(Alom [sec(s) III.A] “The spikes are generated by neurons and can be the sent to any single axon on the chip.” [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential. … The adder (Adder-1) corelet accumulates the input spikes coming from a random spike generator in the present time-tick with the previous time-tick output spikes coming from the second adder named adder-2 close to the output unit. The synaptic weights are assigned after vectorization of the input weight matrix in row major order. QUBO corelets perform the thresholding operation with respect to the value of the membrane potential of neurons. Since the QUBO corelet produces the outputs with respect to individual elements of the weight matrix, therefore the outputs need to be summed up for each column. Adder-2 performs the addition operation on the outputs on the QUBO corelet.”;)
computing, by the second neuron, a cost using a cost function based on the one or more values modified of the one or more variables;
(Alom [sec(s) II] “according to the TrueNorth internal architecture, we can only map the values of the input symmetric matrix Q ϵ Znxn of the QUBO problem as the synaptic weight of neurons which are ranged in signed integers between -128 to 127 on the TrueNorth system [14].” [sec(s) IV.B] “QUBO corelets perform the thresholding operation with respect to the value of the membrane potential of neurons. Since the QUBO corelet produces the outputs with respect to individual elements of the weight matrix, therefore the outputs need to be summed up for each column. Adder-2 performs the addition operation on the outputs on the QUBO corelet.”;)
determining, by a third neuron in the neural network, whether the cost meets a convergence criterion; and
(Alom [sec(s) III] “The membrane potential is compared with the threshold voltage αj every tick. If the membrane potential is greater or equal the desire threshold voltage, the neuron fires a spike and resets the membrane potential to zero.” [sec(s) V.A] “In this implementation, a zero value has been assigned as the threshold of neurons.” [sec(s) V.D] “From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes. This is quite fast in terms of convergence. Furthermore, the QUBO problem can be implemented for a minimization value function. The proposed technique can be easily implemented for minimization approach by changing the sign of individual elements of the input weight matrix.”;)
in response to determining that the cost meets the convergence criterion, transmitting, by the third neuron, a message to the one or more first neurons, the message instructing the one or more first neurons to stop changing the one or more states.
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) IV.A] “In JRNN, the outputs of the neural network are used as inputs, with inputs of tth time-step. Since, the TrueNorth system only deals with spikes as inputs and outputs, we have implemented a spiking form of JRNN here, named Vanilla RNN in Fig 2. According to Fig. 2, the inputs and outputs of the recurrent network are represented with x(t) and y(t) respectively. Here h(t) is used for the outputs of hidden layer and the context inputs of this recurrent network are represented with y(t-1). This is the delayed version of output y(t).” [sec(s) IV.B] “There are two sets of outputs generated using Adder-2. One set of outputs is used for recurrent inputs and another output set is shown as final outputs. The output corelet shows the final spikes as outputs. The internal core structure of the TrueNorth implementation with respect to the different operations are shown in Fig. 4. In this figure, the internal core structures have been shown for better understanding for an 8x8 input weight matrix for the solution of an 8 nodes input graph.” [sec(s) V.D] “From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes.”;)
Regarding claim 2
The combination of Alom teaches claim 1.
Alom teaches
a first neuron comprises a spiking unit, a first unit, and a second unit,
(Alom [fig(s) 1] “IBM’s Neurosynaptic Cognitive TrueNorth Chips (a) TrueNorth multi-chip system (b) a single chip and (c) a zoomed-in internal structure of single core.” [sec(s) III.B] “There are different types of neuron models that have been used in the TrueNorth system. The Leaky Integrate-and-Fire (LIF) neurons are used in this study. The basic operations are (see equations 3-5): 1. synaptic integration, 2. leak integration, 3. threshold, 4. spike firing, and 5. reset. In the general case, the LIF neuron model can be described by the following equations 3-5 [14]. Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.”;)
the spiking unit receives a first input from the first unit and receives a second input from the second unit, and
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential. … The adder (Adder-1) corelet accumulates the input spikes coming from a random spike generator in the present time-tick with the previous time-tick output spikes coming from the second adder named adder-2 close to the output unit. The synaptic weights are assigned after vectorization of the input weight matrix in row major order. QUBO corelets perform the thresholding operation with respect to the value of the membrane potential of neurons. Since the QUBO corelet produces the outputs with respect to individual elements of the weight matrix, therefore the outputs need to be summed up for each column. Adder-2 performs the addition operation on the outputs on the QUBO corelet.”;)
the spiking unit updates a state of the first neuron based on the first input and the second input.
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick. From the architecture’s point of view, each axon is assigned 0 of 3 axon types, which are used as an index into a “s-value” lookup table. The s-value table is unique to each neuron and provides a signed 9-bit integer synaptic strength to the corresponding synapse. The spikes are generated by neurons and can be the sent to any single axon on the chip. Each neuron in a core can be represented with about 23 individual programmable features such as synaptic weight, crossbar weight, threshold, leak, and reset.”;)
Regarding claim 3
The combination of Alom teaches claim 2.
Alom teaches
the first unit computes the first input based on data received from another first neuron, and
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick.”;)
the second input from the second neuron is a prior state of the first neuron.
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick.”;)
Regarding claim 4
The combination of Alom teaches claim 2.
Alom teaches
the first neuron further comprises an additional unit, and
(Alom [fig(s) 1] “IBM’s Neurosynaptic Cognitive TrueNorth Chips (a) TrueNorth multi-chip system (b) a single chip and (c) a zoomed-in internal structure of single core.” [sec(s) III.B] “There are different types of neuron models that have been used in the TrueNorth system. The Leaky Integrate-and-Fire (LIF) neurons are used in this study. The basic operations are (see equations 3-5): 1. synaptic integration, 2. leak integration, 3. threshold, 4. spike firing, and 5. reset. In the general case, the LIF neuron model can be described by the following equations 3-5 [14]. Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) III.A] “The spikes are generated by neurons and can be the sent to any single axon on the chip. Each neuron in a core can be represented with about 23 individual programmable features such as synaptic weight, crossbar weight, threshold, leak, and reset.”;)
the additional unit, based on a message from the third neuron, resets the state of the first neuron to an initialized state of the first neuron.
(Alom [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) IV.A] “In JRNN, the outputs of the neural network are used as inputs, with inputs of tth time-step. Since, the TrueNorth system only deals with spikes as inputs and outputs, we have implemented a spiking form of JRNN here, named Vanilla RNN in Fig 2. According to Fig. 2, the inputs and outputs of the recurrent network are represented with x(t) and y(t) respectively. Here h(t) is used for the outputs of hidden layer and the context inputs of this recurrent network are represented with y(t-1). This is the delayed version of output y(t).” [sec(s) IV.B] “There are two sets of outputs generated using Adder-2. One set of outputs is used for recurrent inputs and another output set is shown as final outputs. The output corelet shows the final spikes as outputs. The internal core structure of the TrueNorth implementation with respect to the different operations are shown in Fig. 4. In this figure, the internal core structures have been shown for better understanding for an 8x8 input weight matrix for the solution of an 8 nodes input graph.”;)
Regarding claim 6
The combination of Alom teaches claim 1.
Alom teaches
wherein the message further instructs the one or more first neurons to send a processing unit the one or more spikes as a solution to the optimization problem.
(Alom [fig(s) 4] “Produce two sets of outputs after addition on QUBO corelet outputs; one set of outputs are used as recurrent inputs and another set of outputs are used as input of output corelet.” [sec(s) IV.B] “There are two sets of outputs generated using Adder-2. One set of outputs is used for recurrent inputs and another output set is shown as final outputs. The output corelet shows the final spikes as outputs.” [sec(s) V.A] “Fig. 6(b) shows the final outputs spikes on the TrueNorth system. The spikes of respective pin numbers show the active node on the input graph. According to the outputs in Fig. 6(b), it is clearly seen that only pins {1, 2, 3, and 6} are active. In this implementation, a zero value has been assigned as the threshold of neurons.” [sec(s) V.C] “The TrueNorth system provides accurate solution set {1, 2, 3, 5,8,11, and 13}, as shown in Fig. 10. (b).” [sec(s) V.D] “From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes. This is quite fast in terms of convergence.”;)
Regarding claim 7
The combination of Alom teaches claim 1.
Alom teaches
in response to determining that the cost fails to meet the convergence criterion, transmitting, by the third neuron, a different message to one or more units in the one or more first neurons, the different message instructing the one or more units in the one or more first neurons to further modify the one or more values of the one or more variables by further changing the one or more states of the one or more first neurons.
(Alom [sec(s) V.D] “From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes. This is quite fast in terms of convergence.” [sec(s) IV.A] “Here h(t) is used for the outputs of hidden layer and the context inputs of this recurrent network are represented with y(t-1). This is the delayed version of output y(t).” [sec(s) IV.B] “The adder (Adder-1) corelet accumulates the input spikes coming from a random spike generator in the present time-tick with the previous time-tick output spikes coming from the second adder named adder-2 close to the output unit. … There are two sets of outputs generated using Adder-2. One set of outputs is used for recurrent inputs and another output set is shown as final outputs. The output corelet shows the final spikes as outputs. The internal core structure of the TrueNorth implementation with respect to the different operations are shown in Fig. 4. In this figure, the internal core structures have been shown for better understanding for an 8x8 input weight matrix for the solution of an 8 nodes input graph.” [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5)” [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick.”;)
Regarding claim 9
The combination of Alom teaches claim 1.
wherein determining whether the cost meets a convergence criterion comprises: (See claim 1)
determining whether the cost is equal to or lower than a target cost.
(Alom [sec(s) I] “QUBO is a mathematical optimization programming problem, the objective of this approach is to find the minimum (or the maximum) value of a quadratic function with a finite number of binary variables [1-3].” [sec(s) III.B] “Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential. The membrane potential is compared with the threshold voltage αj every tick. If the membrane potential is greater or equal the desire threshold voltage, the neuron fires a spike and resets the membrane potential to zero.” [sec(s) V.D] “From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes. This is quite fast in terms of convergence. Furthermore, the QUBO problem can be implemented for a minimization value function. The proposed technique can be easily implemented for minimization approach by changing the sign of individual elements of the input weight matrix.” [sec(s) V.E] “In most of the cases for solving a particular problem, we usually have predefined values for making the decision in most real life problems. In the TrueNorth system, we can easily implement multi-level thresholding after encoding the values according to a neuron’s capacity to -128 to 127 in the present architecture. First, the highest value will be considered as threshold, and the active nodes will be shown as results.”;)
Regarding claim 10
The combination of Alom teaches claim 1.
wherein determining whether the cost meets a convergence criterion comprises: (See claim 1)
Alom teaches
determining whether a number of steps in which the one or more first neurons change the one or more states exceeds a threshold number.
(Alom [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick. From the architecture’s point of view, each axon is assigned 0 of 3 axon types, which are used as an index into a “s-value” lookup table. The s-value table is unique to each neuron and provides a signed 9-bit integer synaptic strength to the corresponding synapse. The spikes are generated by neurons and can be the sent to any single axon on the chip. Each neuron in a core can be represented with about 23 individual programmable features such as synaptic weight, crossbar weight, threshold, leak, and reset.” [sec(s) IV.B] “We have run for 50 and 100 time-ticks in this implementation. The adder (Adder-1) corelet accumulates the input spikes coming from a random spike generator in the present time-tick with the previous time-tick output spikes coming from the second adder named adder-2 close to the output unit.” [sec(s) V.D] “Random inputs are considered for this experiment over a period of 50 time ticks. It is clearly seen that the proposed system provides an accurate solution set for this input graph as shown in Fig. 12 (b). From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes. This is quite fast in terms of convergence. Furthermore, the QUBO problem can be implemented for a minimization value function. The proposed technique can be easily implemented for minimization approach by changing the sign of individual elements of the input weight matrix.”;)
Regarding claim 11
The claim is a computer-readable media claim corresponding to the method claim 1, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 12
The claim is a computer-readable media claim corresponding to the method claim 2, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 13
The claim is a computer-readable media claim corresponding to the method claim 3, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 14
The claim is a computer-readable media claim corresponding to the method claim 4, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 16
The claim is a computer-readable media claim corresponding to the method claim 6, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 17
The claim is a computer-readable media claim corresponding to the method claim 7, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 19
The claim is a system claim corresponding to the method claim 1, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Regarding claim 20
The claim is a system claim corresponding to the method claim 2, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 5, 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Alom et al. (Quadratic Unconstrained Binary Optimization (QUBO) on Neuromorphic Computing System) in view of Brette et al. (Adaptive Exponential Integrate-and-Fire Model as an Effective Description of Neuronal Activity)
Regarding claim 5
The combination of Alom teaches claim 2.
Alom teaches
the first neuron further comprises an additional unit,
(Alom [fig(s) 1] “IBM’s Neurosynaptic Cognitive TrueNorth Chips (a) TrueNorth multi-chip system (b) a single chip and (c) a zoomed-in internal structure of single core.” [sec(s) III.B] “There are different types of neuron models that have been used in the TrueNorth system. The Leaky Integrate-and-Fire (LIF) neurons are used in this study. The basic operations are (see equations 3-5): 1. synaptic integration, 2. leak integration, 3. threshold, 4. spike firing, and 5. reset. In the general case, the LIF neuron model can be described by the following equations 3-5 [14]. Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) III.A] “The spikes are generated by neurons and can be the sent to any single axon on the chip. Each neuron in a core can be represented with about 23 individual programmable features such as synaptic weight, crossbar weight, threshold, leak, and reset.”;)
the additional unit computes a [time-weighted average] of states of the first neuron, and
(Alom [sec(s) III.B] “1. synaptic integration, 2. leak integration, 3. threshold, 4. spike firing, and 5. reset. In the general case, the LIF neuron model can be described by the following equations 3-5 [14]. Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential.” [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick.”;)
the spiking unit sends out a spike encoding the state of the first neuron based on a determination that the state of the first neuron is equal to or greater than the [time-weighted average].
(Alom [sec(s) III.B] “1. synaptic integration, 2. leak integration, 3. threshold, 4. spike firing, and 5. reset. In the general case, the LIF neuron model can be described by the following equations 3-5 [14]. Synaptic integration:
PNG
media_image1.png
80
890
media_image1.png
Greyscale
(3) Leak integration:
PNG
media_image2.png
76
499
media_image2.png
Greyscale
(4) Threshold, fire, and reset
PNG
media_image3.png
290
494
media_image3.png
Greyscale
(5) The parameter Vj(t) represents for the summation of the membrane potential of the jth neuron in the tth timestep, and Vj(t-1) is the sum of the membrane potential of the prior time-step. Xi(t), and si are the synaptic inputs as sum of spike inputs in the current time-step and signed synaptic weights respectively. The leak value λj is subtracted in every time-step from the membrane potential. The membrane potential is compared with the threshold voltage αj every tick. If the membrane potential is greater or equal the desire threshold voltage, the neuron fires a spike and resets the membrane potential to zero.” [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick.”;)
However, the combination of Alom does not appear to explicitly teach:
the additional unit computes a [time-weighted average] of states of the first neuron, and
the spiking unit sends out a spike encoding the state of the first neuron based on a determination that the state of the first neuron is equal to or greater than the [time-weighted average].
Brette teaches
the additional unit computes a time-weighted average of states of the first neuron, and
the spiking unit sends out a spike encoding the state of the first neuron based on a determination that the state of the first neuron is equal to or greater than the time-weighted average.
(Brette [sec(s) METHODS] “The adaptation current w is defined by
PNG
media_image4.png
104
447
media_image4.png
Greyscale
(3) where τw is the time constant and a represents the level of subthreshold adaptation. At each firing time, the variable w is increased by an amount b, which accounts for spike-triggered adaptation. Izhikevich (2003) showed that when the quadratic integrate-and-fire model is augmented by this adaptation current, it can be tuned to reproduce qualitatively all major classes of neurons, as defined electrophysiologically in vitro (e.g., regular spiking, intrinsic bursting, fast spiking). For example, high reset values (Vr > VT) induce bursting; large values of b give strong spike-frequency adaptation; high values of the parameter a yield subthreshold oscillations, and medium values cause overshoots in response to current pulses.”; Note that the differential equation acts as a low-pass filter, maintaining a time-weighted average.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the system of Alom with the time-weighted average of Brette.
One of ordinary skill in the art would have been motived to combine in order to show that an exponential integrate-and-fire model with adaptation can accurately predict the spike trains of a detailed regular spiking neuron model driven by realistic, conductance-based, synaptic inputs.
(Davies [sec(s) DISCUSSION] “Our results show that an exponential integrate-and-fire model with adaptation can accurately predict the spike trains of a detailed regular spiking neuron model driven by realistic, conductance-based, synaptic inputs. The aEIF model combines an exponential spike mechanism (Fourcaud-Trocme et al. 2003) with an adaptation equation with reset (Izhikevich 2003). It predicted, on average, 96% of the spikes with 2-ms precision, improving significantly on the leaky integrate-and fire model with adaptation (88%).”)
Regarding claim 15
The claim is a computer-readable media claim corresponding to the method claim 5, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Claim(s) 8, 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Alom et al. (Quadratic Unconstrained Binary Optimization (QUBO) on Neuromorphic Computing System) in view of Davies et al. (Advancing Neuromorphic Computing With Loihi: A Survey of Results and Outlook)
Regarding claim 8
The combination of Alom teaches claim 1.
Alom teaches
determining, by a fourth neuron in the neural network, whether a [stalling] period threshold is reached based on a spike from the third neuron; and
(Alom [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick. From the architecture’s point of view, each axon is assigned 0 of 3 axon types, which are used as an index into a “s-value” lookup table. The s-value table is unique to each neuron and provides a signed 9-bit integer synaptic strength to the corresponding synapse. The spikes are generated by neurons and can be the sent to any single axon on the chip. Each neuron in a core can be represented with about 23 individual programmable features such as synaptic weight, crossbar weight, threshold, leak, and reset.” [sec(s) III.B] “The leak value λj is subtracted in every time-step from the membrane potential. The membrane potential is compared with the threshold voltage αj every tick. If the membrane potential is greater or equal the desire threshold voltage, the neuron fires a spike and resets the membrane potential to zero.” [sec(s) V.D] “Random inputs are considered for this experiment over a period of 50 time ticks. It is clearly seen that the proposed system provides an accurate solution set for this input graph as shown in Fig. 12 (b). From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes.”;)
after determining that [stalling] period threshold is reached, instructing, by the fourth neuron, the third neuron to transmit a different message to the one or more first neurons, the different message instructing the one or more first neurons to change the one or more modified values of the one or more variables back to the one or more values of the one or more variables.
(Alom [sec(s) III.A] “In the TrueNorth architecture, each neuron’s and synapse’s states are updated in every millisecond. This duration is called a tick. From the architecture’s point of view, each axon is assigned 0 of 3 axon types, which are used as an index into a “s-value” lookup table. The s-value table is unique to each neuron and provides a signed 9-bit integer synaptic strength to the corresponding synapse. The spikes are generated by neurons and can be the sent to any single axon on the chip. Each neuron in a core can be represented with about 23 individual programmable features such as synaptic weight, crossbar weight, threshold, leak, and reset.” [sec(s) IV.A] “In JRNN, the outputs of the neural network are used as inputs, with inputs of tth time-step. Since, the TrueNorth system only deals with spikes as inputs and outputs, we have implemented a spiking form of JRNN here, named Vanilla RNN in Fig 2. According to Fig. 2, the inputs and outputs of the recurrent network are represented with x(t) and y(t) respectively. Here h(t) is used for the outputs of hidden layer and the context inputs of this recurrent network are represented with y(t-1). This is the delayed version of output y(t). We have considered the weight matrices of WIH, WHO, and WOI for inputs to hidden, hidden to outputs, and outputs to inputs respectively. A recurrent neural network can then be expressed using eq. 6 and 7.
PNG
media_image5.png
129
742
media_image5.png
Greyscale
” [sec(s) IV.B] “We have run for 50 and 100 time-ticks in this implementation. The adder (Adder-1) corelet accumulates the input spikes coming from a random spike generator in the present time-tick with the previous time-tick output spikes coming from the second adder named adder-2 close to the output unit.” [sec(s) V.D] “Random inputs are considered for this experiment over a period of 50 time ticks. It is clearly seen that the proposed system provides an accurate solution set for this input graph as shown in Fig. 12 (b). From the above experimental results, it is clearly observed that the solutions are converged within 10 to 20 time-tick depending upon the complexity of graph as well as the pattern of random input spikes.”;)
However, the combination of Alom does not appear to explicitly teach:
determining, by a fourth neuron in the neural network, whether a [stalling] period threshold is reached based on a spike from the third neuron; and
after determining that [stalling] period threshold is reached, instructing, by the fourth neuron, the third neuron to transmit a different message to the one or more first neurons.
Davies teaches
determining, by a fourth neuron in the neural network, whether a stalling period threshold is reached based on a spike from the third neuron; and
after determining that stalling period threshold is reached, instructing, by the fourth neuron, the third neuron to transmit a different message to the one or more first neurons.
(Davies [sec(s) IV.A] “NxSDK provides a compiler for convolutional LCA net works that exploit weight sharing to make efficient use of on-chip memory and support networks spanning mil lions of feature neurons. Fig. 4 compares LCA on Loihi4 to a CPU4 running FISTA [54], the leading conventional algorithm. Both algorithms generate sparse codes of input images5 by solving the same LASSO problem to the same quality of the solution, as measured by the LASSO objective function. Loihi LCA objective values typically saturate at ∼1% of the optimal value, which sets the convergence threshold of the evaluations. This is approximate but sufficient for many applications.6 … Given sufficiently long run times, FISTA can solve LASSO to much lower objective values than LCA on Loihi. We also evaluated least angle regression (LARS) [56] and found that it performs worse than FISTA for our problem’s sparsity levels and approximate convergence thresholds.”;)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the system of Alom with the stalling of Davies.
One of ordinary skill in the art would have been motived to combine in order to allow the most promising properties and pressing challenges to be prioritized appropriately.
(Davies [sec(s) I] “Demonstrations that appear impressive can often obscure important caveats. We instead have focused on the fundamentals, so the full scope of the technology can be understood—both strengths and weaknesses. This allows the most promising properties and pressing challenges to be prioritized appropriately.”)
Regarding claim 18
The claim is a computer-readable media claim corresponding to the method claim 8, and is directed to largely the same subject matter. Thus, it is rejected for the same reasons as given in the rejections of the method claim.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEHWAN KIM whose telephone number is (571)270-7409. The examiner can normally be reached Mon - Fri 9:00 AM - 5:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Michael J Huntley can be reached on (303) 297-4307. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SEHWAN KIM/Examiner, Art Unit 2129
Prosecution Insights