DETAILED ACTION
This action is responsive to the Request for Continued Examination filed on 04/22/2026. Claims 1-4 and 6-9 are pending in the case. Claim(s) 5 has been cancelled. Claims 1, 8, and 9 are independent claims. Claims 1, 6, 8 and 9 are amended.
Priority
Foreign Priority is acknowledged through foreign application dated 08/04/2021
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 04/22/2026 has been entered.
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 .
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 (i.e., changing from AIA to pre-AIA ) 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.
Claim Rejections - 35 USC § 112
Claims 1, 8 and 9 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
The limitation edges are drawn with an identical probability based on a second probability in claims 1, 8 and 9 is relative which renders the claim indefinite as the term is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. The limitation is rejected as edges are drawn with an identical probability based on a second probability does not clearly define what the identical probability is identical to.
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 1-9 rejected under 35 U.S.C. 101 because the claims are directed to an abstract idea/mental process.
Regarding claim 1:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites randomly drawing a multitude of subgraphs by the directed graph as a function of the respective variables which under the broadest reasonable interpretation, covers performance of the limitation in the mind with aid of pencil and paper. The limitations encompass a user randomly drawing a graph with respect to variables. See 2106.04.(a)(2).III.C.
The claim recites in which edges are drawn as a function of the respective variables assigned to the edges based on the first probability and edges are drawn with an identical probability based on a second probability which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites wherein a sum of the first probability and the second probability is equal to one which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites the respective variables being changed in the directed graph as a function of a distribution of values of the respective variables… wherein the change of the respective variables takes place as a function of a first probability which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user changing values based on a distribution. See 2106.04.(a)(2).III.C.
The claim recites and drawing a last subgraph, as a function of the adapted respective variables which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user randomly selecting a subgraph. See 2106.04.(a)(2).III.C.
The claim recites during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
The claim recites providing a directed graph including(recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)) which under the broadest reasonable interpretation, covers performance of the limitation in the mind with the aid of pen and paper. The limitations encompass a user drawing out nodes and connect them. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
computer-implemented(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
one or multiple input and output nodes, which are connected via a multitude of edges and nodes, a respective variable being assigned to each respective edge of the edges, which characterizes a probability with which the respective edge is drawn (merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs, wherein … creating the machine learning system corresponding to the last subgraph (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized(Training by adjusting variables/parameters to optimize a function is Insignificant Extra-Solution Activity (see MPEP §2106.05(g))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) and (c) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
Additional elements (b) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
Additional element (d) recites a well understood and conventional practice training by adjusting variables/parameters to optimize a function quoted from US20210073675A1 ([0003], “Building and training models using machine learning generally involves learning a set of parameters of the model by adjusting the parameters to minimize a loss function”
The additional element(s) (a) (b) (c) and (d) in the claim do/does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 2:
The rejection of claim 1 is incorporated and further claim recites further additional
elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein, when a measure of the distribution of the values of the respective variables relative to a predefined target measure of a target distribution is greater which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
The claim recites the respective variables are changed in such a way that edges having an essentially equal probability are drawn which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user deciding probability for a set of a data/a probability distribution. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 3:
The rejection of claim 1 is incorporated and further claim recites further additional
elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein the change of the respective variables takes place as a function of an entropy of the directed graph, and a number of training steps which have already been carried out which, under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user changing variables based on judging a variable and how far along the training process is. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 4:
The rejection of claim 3 is incorporated and further claim recites further additional
elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites when the entropy is greater than a predefined target entropy, which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
The claim recites a parameter by which the respective variables are changed is changed in such a way that it changes values of the respective variables so that the probability distribution characterizing the respective variables has a lesser similarity to a uniform distribution, which, under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user choosing a probability distribution that is less uniform. See 2106.04.(a)(2).III.C.
The claim recites and when the ascertained entropy is smaller than the predefined target entropy, which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
The claim recites the parameter is changed in such a way that it changes values of the respective variables, so that the probability distribution characterizing the respective variables characterizes a uniform distribution which, under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user choosing a probability distribution that is more uniform. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 6:
The rejection of claim 1 is incorporated and further claim recites further additional
elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein the change of the respective variables is further performed based on a temperature scaling which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 7:
The rejection of claim 6 is incorporated and further claim recites further additional
elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein, during the temperature scaling, the respective variables are scaled as a function of a temperature which is changed as a function of the distribution of the values of the respective variables which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 8:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites randomly drawing a multitude of subgraphs by the directed graph as a function of the respective variables which under the broadest reasonable interpretation, covers performance of the limitation in the mind with aid of pencil and paper. The limitations encompass a user randomly drawing a graph with respect to variables. See 2106.04.(a)(2).III.C.
The claim recites in which edges are drawn as a function of the respective variables assigned to the edges based on the first probability and edges are drawn with an identical probability based on a second probability which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites wherein a sum of the first probability and the second probability is equal to one which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites the respective variables being changed in the directed graph as a function of a distribution of values of the respective variables… wherein the change of the respective variables takes place as a function of a first probability which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user changing values based on a distribution. See 2106.04.(a)(2).III.C.
The claim recites and drawing a last subgraph, as a function of the adapted respective variables which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user randomly selecting a subgraph. See 2106.04.(a)(2).III.C.
The claim recites during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
The claim recites providing a directed graph including(recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)) which under the broadest reasonable interpretation, covers performance of the limitation in the mind with the aid of pen and paper. The limitations encompass a user drawing out nodes and connect them. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
A non-transitory machine-readable memory element on which is stored a computer program for creating a machine learning system, the computer program, when executed by a computer, causing the computer to perform the following(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
one or multiple input and output nodes, which are connected via a multitude of edges and nodes, a respective variable being assigned to each respective edge of the edges, which characterizes a probability with which the respective edge is drawn (merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs, wherein … creating the machine learning system corresponding to the last subgraph (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized(Training by adjusting variables/parameters to optimize a function is Insignificant Extra-Solution Activity (see MPEP §2106.05(g))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) and (c) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
Additional elements (b) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
Additional element (d) recites a well understood and conventional practice training by adjusting variables/parameters to optimize a function quoted from US20210073675A1 ([0003], “Building and training models using machine learning generally involves learning a set of parameters of the model by adjusting the parameters to minimize a loss function”
The additional element(s) (a) (b) (c) and (d) in the claim do/does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 9:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites randomly drawing a multitude of subgraphs by the directed graph as a function of the respective variables which under the broadest reasonable interpretation, covers performance of the limitation in the mind with aid of pencil and paper. The limitations encompass a user randomly drawing a graph with respect to variables. See 2106.04.(a)(2).III.C.
The claim recites in which edges are drawn as a function of the respective variables assigned to the edges based on the first probability and edges are drawn with an identical probability based on a second probability which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites wherein a sum of the first probability and the second probability is equal to one which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites the respective variables being changed in the directed graph as a function of a distribution of values of the respective variables… wherein the change of the respective variables takes place as a function of a first probability which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user changing values based on a distribution. See 2106.04.(a)(2).III.C.
The claim recites and drawing a last subgraph, as a function of the adapted respective variables which under the broadest reasonable interpretation, covers performance of the limitation in the mind. The limitations encompass a user randomly selecting a subgraph. See 2106.04.(a)(2).III.C.
The claim recites during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
The claim recites providing a directed graph including(recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)) which under the broadest reasonable interpretation, covers performance of the limitation in the mind with the aid of pen and paper. The limitations encompass a user drawing out nodes and connect them. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
device configured to create a machine learning system(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
one or multiple input and output nodes, which are connected via a multitude of edges and nodes, a respective variable being assigned to each respective edge of the edges, which characterizes a probability with which the respective edge is drawn (merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
train a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs, wherein … create the machine learning system corresponding to the last subgraph (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized(Training by adjusting variables/parameters to optimize a function is Insignificant Extra-Solution Activity (see MPEP §2106.05(g))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) and (c) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
Additional elements (b) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
Additional element (d) recites a well understood and conventional practice training by adjusting variables/parameters to optimize a function quoted from US20210073675A1 ([0003], “Building and training models using machine learning generally involves learning a set of parameters of the model by adjusting the parameters to minimize a loss function”
The additional element(s) (a) (b) (c) and (d) in the claim do/does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
.
Claim Rejections - 35 USC § 103
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1 and 5-9 is/are rejected under 35 U.S.C. 103 as being obvious over Xie et al. (SNAS: STOCHASTIC NEURAL ARCHITECTURE SEARCH, henceforth known as Xie) in view of Baker et al. (DESIGNING NEURAL NETWORK ARCHITECTURES USING REINFORCEMENT LEARNING, henceforth known as Baker)
Regarding claim 1:
Xie discloses a computer-implemented method(Xie, Page 6, Footnote 2, “All the experiments were performed using NVIDIA TITAN Xp GPUs”) for creating a machine learning system(Xie, Page 5, Paragraph 7, “First, SNAS is applied to search for convolutional cells in a small parent network on CIFAR-10 and we choose the best cells based on their search validation accuracy. Then, a larger network is constructed by stacking the learned cells (child graphs) and is retrained on CIFAR-10 to compare the performance of SNAS with other state-of-the-art methods.”)
Xie discloses providing a directed graph including one or multiple input and output nodes, which are connected via a multitude of edges and nodes(Xie, Figure 1, where Figure 1 shows the nodes 0, 1, 2, and 3, multiple edges connecting the nodes and multiple input and output nodes), a respective variable being assigned to each respective edge of the edges(Xie, Page 2, Figure 1, “Sampled from p(Z), Z is a matrix whose rows Zi,j are one-hot random variable vectors…” and Xie, Page 3, Paragraph 2, “As shown in the left of Figure 1, the search space, i.e. a cell, is represented using a directed acyclic graph (DAG), which is called parent graph. Nodes xi in this DAG represent latent representation, whose dimensions are simply ignored to avoid abuse of notations. In convolutional networks, they are feature maps. Edges (i, j) represent information flows and possible operations Oi,j to be selected between two nodes xi and xj” where Zi,j is a variable representing architecture distribution over edges is considered a respective variable being assigned to each respective edge which characterizes a probability with which the respective edge is drawn), which characterizes a probability with which the respective edge is drawn(Xie, Page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph” where Zi,j is the sampled selector outcome and corresponds to whether an edge is drawn after probabilistic selection which corresponds to a probability to which a respective edge is drawn)
Xie discloses randomly drawing a multitude of subgraphs by the directed graph(Xie, Page 3, Paragraph 1 and Equation 2, “…the search space is represented with a set of one-hot random variables from a fully factorizable joint distribution, multiplied as a mask to select operations in the graph” where the parent DAG has edges (i, j) with multiple operations and one operation is sample per edge to form a child graph) as a function of the respective variables(Xie, Page 3, Equation 2, where Zij child graph has intermediate nodes indicating what operation is to be used and the stochastic selection is considered drawing random subgraphs based on the function of respective variables), the respective variables being changed in the directed graph as a function of a distribution of values of the respective variables(Xie, Page 3, Paragraph 5, “In SNAS, we simply assume that p(Z) is fully factorizable, whose factors are parameterized with α and learnt along with operation parameters θ” where the distribution of p(Z) that is parameterized with α with operation parameters θ being updated using gradient descent corresponds to updating respective variables of the original graph), wherein the change of the respective variables takes place function of a first probability(Page 4, Equation 5 with Paragraph 2 “it is proved that p(limλ→0 Zki,j = 1) = αki,j/(
∑
l
=
0
n
α
i
,
j
k
j
)”, where αi,j corresponds to a first probability as it the distribution used to generate Zi,j and the parent DAG has edges (i, j) with multiple operations and one operation is sampling per edge to form a child graph using the concrete distribution as a continuous relaxation of pα(Z)) in which edges are drawn as a function of the respective variables assigned to the edges(Xie, Page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph”) based on the first probability(Xie, Page 2, Paragraph 1, “Sampling from this search space is made differentiable by relaxing the architecture distribution with concrete distribution” and Page 4, Equation 5 with Paragraph 1-2, where edges being realized as a function of the variables assigned to them and those variables being sampled according to probability distribution pα(Z) and the exploratory stochastic sampling for per-edge probability of the global exploratory distribution pα(Z) corresponds edges are drawn as a function of the respective variable the first probability)
Xie discloses training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs(Xie, Page 6, Paragraph 11, “…we follow this assumption in evaluation stage, stacking more cells (child graphs) to build a deeper network. This network is trained from scratch…”), wherein during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized(Xie, Page 4, Equations 6, where Equation 6 provides gradients with respect to parameters(θ) and architecture parameters(α) shows adaptation during training based on respective variables) and drawing a last subgraph, as a function of the adapted respective variables, and creating the machine learning system corresponding to the last subgraph(Xie, Page 5, Paragraph 6, “First, SNAS is applied to search for convolutional cells in a small parent network on CIFAR-10 and we choose the best cells based on their search validation accuracy. Then, a larger network is constructed by stacking the learned cells (child graphs) and is retrained on CIFAR-10 to compare the performance of SNAS with other state-of-the-art methods.” Where stacking learning child graphs is considered drawing a subgraph with respect to adapted respective variables and creating a machine learning system corresponding to a subgraph)
Xie does not disclose, however Baker discloses edges are drawn…based on the first probability(Baker, Page 3, Paragraph 6, “We choose the behavior distribution using an-greedy strategy…With this strategy, a random action is taken with probability ε and the greedy action… is chosen with probability 1 − ε” where the selection greedy exploitation(1- ε) corresponds to a first probability and the behavior distribution corresponds to traversing/drawing edges as the states and actions/transitions as described in Baker form a graph(See also Baker, Page 3, Paragraph 7, “We model the layer selection process as a Markov Decision Process with the assumption that a well-performing layer”)) and edges are drawn with an identical probability based on a second probability(Baker, Page 5, Paragraph 4, “At ε = 1.0, the agent samples CNN architecture with a random walk along a uniformly weighted Markov chain” where a random along a uniformly weighted Markov chain corresponds to edges are drawn with an identical probability based on a second probability as a uniformly weighted Markov chain transitions out of a given state choose the next edge with equal probability and the likelihood of edges being drawn with an identical probability being dictated by probability ε corresponds being based on a second probability), wherein a sum of the first probability and the second probability is equal to one(Baker, Page 3, Paragraph 6, “We choose the behavior distribution using an-greedy strategy…With this strategy, a random action is taken with probability ε and the greedy action… is chosen with probability 1 − ε” where the probability of a random exploration(ε) and greedy exploitation(1- ε) summing to 1 corresponds with wherein a sum of the first probability and the second probability is equal to one probability)
References Xie and Baker are analogous art because they are from the same field of endeavor of automated neural network architecture design
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Xie and Baker before him or her, to modify SNAS’s exploratory stochastic edge sampling of Xie to include the epsilon greedy exploration/exploitation of Baker as both are mechanisms control stochastic exploration and the epsilon greedy method has the capacity to handle large or continuous state spaces. The suggestion/motivation for doing so would have been “The number of possible choices makes the design space of CNN architectures extremely large and hence, infeasible for an exhaustive manual search”(Baker, Page 1, Paragraph 1) and “However, in the case of large or continuous state spaces, the-greedy strategy of learning has been empirically shown to converge” ”(Baker, Page 2, Paragraph 4)
Regarding claim 6:
The rejection of claim 1 with prior art Xie-Baker is incorporated and further:
Xie further discloses wherein the change of the respective variables is further performed based on temperature scaling(Xie, Page 4, Equation 5 and Paragraph 2, “λ is the temperature of the SoftMax, which is steadily annealed to be close to zero in SNAS” and where Equation 5 shows that the sampling using temperature λ with architecture distribution during training is corresponds to a change in respective variable based on using temperature scaling)
Regarding claim 7:
The rejection of claim 6 with prior art Xie-Baker is incorporated and further:
Xie further discloses wherein, during the temperature scaling, the respective variables are scaled as a function of a temperature(Xie, Page 4, Paragraph 2, “λ is the temperature of the SoftMax, which is steadily annealed to be close to zero in SNAS”) which is changed as a function of the distribution of the values of the respective variables(Equation 5, where Zki,j uses 1/λ in SoftMax and the steady annealing of λ during training is tied to the evolving architecture distribution which is considered temperature changing as a function of the distribution of values of the respective variables as early α values are uncertain and will have a higher λ values and, as the α distribution sharpens from training, λ is decreased to encourage hard selections reflecting a learned distribution)
Regarding claim 8:
Xie discloses a non-transitory machine-readable memory element on which is stored a (Xie, Page 6, Footnote 2, “All the experiments were performed using NVIDIA TITAN Xp GPUs”) computer program for creating a machine learning system(Xie, Page 5, Paragraph 7, “First, SNAS is applied to search for convolutional cells in a small parent network on CIFAR-10 and we choose the best cells based on their search validation accuracy. Then, a larger network is constructed by stacking the learned cells (child graphs) and is retrained on CIFAR-10 to compare the performance of SNAS with other state-of-the-art methods.”)
Xie discloses providing a directed graph including one or multiple input and output nodes, which are connected via a multitude of edges and nodes(Xie, Figure 1, where Figure 1 shows the nodes 0, 1, 2, and 3, multiple edges connecting the nodes and multiple input and output nodes), a respective variable being assigned to each respective edge of the edges(Xie, Page 2, Figure 1, “Sampled from p(Z), Z is a matrix whose rows Zi,j are one-hot random variable vectors…” and Xie, Page 3, Paragraph 2, “As shown in the left of Figure 1, the search space, i.e. a cell, is represented using a directed acyclic graph (DAG), which is called parent graph. Nodes xi in this DAG represent latent representation, whose dimensions are simply ignored to avoid abuse of notations. In convolutional networks, they are feature maps. Edges (i, j) represent information flows and possible operations Oi,j to be selected between two nodes xi and xj” where Zi,j is a variable representing architecture distribution over edges is considered a respective variable being assigned to each respective edge which characterizes a probability with which the respective edge is drawn), which characterizes a probability with which the respective edge is drawn(Xie, Page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph” where Zi,j is the sampled selector outcome and corresponds to whether an edge is drawn after probabilistic selection which corresponds to a probability to which a respective edge is drawn)
Xie discloses randomly drawing a multitude of subgraphs by the directed graph(Xie, Page 3, Paragraph 1 and Equation 2, “…the search space is represented with a set of one-hot random variables from a fully factorizable joint distribution, multiplied as a mask to select operations in the graph” where the parent DAG has edges (i, j) with multiple operations and one operation is sample per edge to form a child graph) as a function of the respective variables(Xie, Page 3, Equation 2, where Zij child graph has intermediate nodes indicating what operation is to be used and the stochastic selection is considered drawing random subgraphs based on the function of respective variables), the respective variables being changed in the directed graph as a function of a distribution of values of the respective variables(Xie, Page 3, Paragraph 5, “In SNAS, we simply assume that p(Z) is fully factorizable, whose factors are parameterized with α and learnt along with operation parameters θ” where the distribution of p(Z) that is parameterized with α with operation parameters θ being updated using gradient descent corresponds to updating respective variables of the original graph), wherein the change of the respective variables takes place function of a first probability(Page 4, Equation 5 with Paragraph 2 “it is proved that p(limλ→0 Zki,j = 1) = αki,j/(
∑
l
=
0
n
α
i
,
j
k
j
)”, where αi,j corresponds to a first probability as it the distribution used to generate Zi,j and the parent DAG has edges (i, j) with multiple operations and one operation is sampling per edge to form a child graph using the concrete distribution as a continuous relaxation of pα(Z)) in which edges are drawn as a function of the respective variables assigned to the edges(Xie, Page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph”) based on the first probability(Xie, Page 2, Paragraph 1, “Sampling from this search space is made differentiable by relaxing the architecture distribution with concrete distribution” and Page 4, Equation 5 with Paragraph 1-2, where edges being realized as a function of the variables assigned to them and those variables being sampled according to probability distribution pα(Z) and the exploratory stochastic sampling for per-edge probability of the global exploratory distribution pα(Z) corresponds edges are drawn as a function of the respective variable the first probability)
Xie discloses training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs(Xie, Page 6, Paragraph 11, “…we follow this assumption in evaluation stage, stacking more cells (child graphs) to build a deeper network. This network is trained from scratch…”), wherein during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized(Xie, Page 4, Equations 6, where Equation 6 provides gradients with respect to parameters(θ) and architecture parameters(α) shows adaptation during training based on respective variables) and drawing a last subgraph, as a function of the adapted respective variables, and creating the machine learning system corresponding to the last subgraph(Xie, Page 5, Paragraph 6, “First, SNAS is applied to search for convolutional cells in a small parent network on CIFAR-10 and we choose the best cells based on their search validation accuracy. Then, a larger network is constructed by stacking the learned cells (child graphs) and is retrained on CIFAR-10 to compare the performance of SNAS with other state-of-the-art methods.” Where stacking learning child graphs is considered drawing a subgraph with respect to adapted respective variables and creating a machine learning system corresponding to a subgraph)
Xie does not disclose, however Baker discloses edges are drawn…based on the first probability(Baker, Page 3, Paragraph 6, “We choose the behavior distribution using an-greedy strategy…With this strategy, a random action is taken with probability ε and the greedy action… is chosen with probability 1 − ε” where the selection greedy exploitation(1- ε) corresponds to a first probability and the behavior distribution corresponds to traversing/drawing edges as the states and actions/transitions as described in Baker form a graph(See also Baker, Page 3, Paragraph 7, “We model the layer selection process as a Markov Decision Process with the assumption that a well-performing layer”)) and edges are drawn with an identical probability based on a second probability(Baker, Page 5, Paragraph 4, “At ε = 1.0, the agent samples CNN architecture with a random walk along a uniformly weighted Markov chain” where a random along a uniformly weighted Markov chain corresponds to edges are drawn with an identical probability based on a second probability as a uniformly weighted Markov chain transitions out of a given state choose the next edge with equal probability and the likelihood of edges being drawn with an identical probability being dictated by probability ε corresponds being based on a second probability), wherein a sum of the first probability and the second probability is equal to one(Baker, Page 3, Paragraph 6, “We choose the behavior distribution using an-greedy strategy…With this strategy, a random action is taken with probability ε and the greedy action… is chosen with probability 1 − ε” where the probability of a random exploration(ε) and greedy exploitation(1- ε) summing to 1 corresponds with wherein a sum of the first probability and the second probability is equal to one probability)
References Xie and Baker are analogous art because they are from the same field of endeavor of automated neural network architecture design
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Xie and Baker before him or her, to modify SNAS’s exploratory stochastic edge sampling of Xie to include the epsilon greedy exploration/exploitation of Baker as both are mechanisms to control stochastic exploration and the epsilon greedy method has the capacity to handle large or continuous state spaces. The suggestion/motivation for doing so would have been “The number of possible choices makes the design space of CNN architectures extremely large and hence, infeasible for an exhaustive manual search”(Baker, Page 1, Paragraph 1) and “However, in the case of large or continuous state spaces, the-greedy strategy of learning has been empirically shown to converge” ”(Baker, Page 2, Paragraph 4)
Regarding claim 9:
Xie discloses a device configured to (Xie, Page 6, Footnote 2, “All the experiments were performed using NVIDIA TITAN Xp GPUs”) create a machine learning system (Xie, Page 5, Paragraph 7, “First, SNAS is applied to search for convolutional cells in a small parent network on CIFAR-10 and we choose the best cells based on their search validation accuracy. Then, a larger network is constructed by stacking the learned cells (child graphs) and is retrained on CIFAR-10 to compare the performance of SNAS with other state-of-the-art methods.”)
Xie discloses providing a directed graph including one or multiple input and output nodes, which are connected via a multitude of edges and nodes(Xie, Figure 1, where Figure 1 shows the nodes 0, 1, 2, and 3, multiple edges connecting the nodes and multiple input and output nodes), a respective variable being assigned to each respective edge of the edges(Xie, Page 2, Figure 1, “Sampled from p(Z), Z is a matrix whose rows Zi,j are one-hot random variable vectors…” and Xie, Page 3, Paragraph 2, “As shown in the left of Figure 1, the search space, i.e. a cell, is represented using a directed acyclic graph (DAG), which is called parent graph. Nodes xi in this DAG represent latent representation, whose dimensions are simply ignored to avoid abuse of notations. In convolutional networks, they are feature maps. Edges (i, j) represent information flows and possible operations Oi,j to be selected between two nodes xi and xj” where Zi,j is a variable representing architecture distribution over edges is considered a respective variable being assigned to each respective edge which characterizes a probability with which the respective edge is drawn), which characterizes a probability with which the respective edge is drawn(Xie, Page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph” where Zi,j is the sampled selector outcome and corresponds to whether an edge is drawn after probabilistic selection which corresponds to a probability to which a respective edge is drawn)
Xie discloses randomly drawing a multitude of subgraphs by the directed graph(Xie, Page 3, Paragraph 1 and Equation 2, “…the search space is represented with a set of one-hot random variables from a fully factorizable joint distribution, multiplied as a mask to select operations in the graph” where the parent DAG has edges (i, j) with multiple operations and one operation is sample per edge to form a child graph) as a function of the respective variables(Xie, Page 3, Equation 2, where Zij child graph has intermediate nodes indicating what operation is to be used and the stochastic selection is considered drawing random subgraphs based on the function of respective variables), the respective variables being changed in the directed graph as a function of a distribution of values of the respective variables(Xie, Page 3, Paragraph 5, “In SNAS, we simply assume that p(Z) is fully factorizable, whose factors are parameterized with α and learnt along with operation parameters θ” where the distribution of p(Z) that is parameterized with α with operation parameters θ being updated using gradient descent corresponds to updating respective variables of the original graph), wherein the change of the respective variables takes place function of a first probability(Page 4, Equation 5 with Paragraph 2 “it is proved that p(limλ→0 Zki,j = 1) = αki,j/(
∑
l
=
0
n
α
i
,
j
k
j
)”, where αi,j corresponds to a first probability as it the distribution used to generate Zi,j and the parent DAG has edges (i, j) with multiple operations and one operation is sampling per edge to form a child graph using the concrete distribution as a continuous relaxation of pα(Z)) in which edges are drawn as a function of the respective variables assigned to the edges(Xie, Page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph”) based on the first probability(Xie, Page 2, Paragraph 1, “Sampling from this search space is made differentiable by relaxing the architecture distribution with concrete distribution” and Page 4, Equation 5 with Paragraph 1-2, where edges being realized as a function of the variables assigned to them and those variables being sampled according to probability distribution pα(Z) and the exploratory stochastic sampling for per-edge probability of the global exploratory distribution pα(Z) corresponds edges are drawn as a function of the respective variable the first probability)
Xie discloses training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs(Xie, Page 6, Paragraph 11, “…we follow this assumption in evaluation stage, stacking more cells (child graphs) to build a deeper network. This network is trained from scratch…”), wherein during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized(Xie, Page 4, Equations 6, where Equation 6 provides gradients with respect to parameters(θ) and architecture parameters(α) shows adaptation during training based on respective variables) and drawing a last subgraph, as a function of the adapted respective variables, and creating the machine learning system corresponding to the last subgraph(Xie, Page 5, Paragraph 6, “First, SNAS is applied to search for convolutional cells in a small parent network on CIFAR-10 and we choose the best cells based on their search validation accuracy. Then, a larger network is constructed by stacking the learned cells (child graphs) and is retrained on CIFAR-10 to compare the performance of SNAS with other state-of-the-art methods.” Where stacking learning child graphs is considered drawing a subgraph with respect to adapted respective variables and creating a machine learning system corresponding to a subgraph)
Xie does not disclose, however Baker discloses edges are drawn…based on the first probability(Baker, Page 3, Paragraph 6, “We choose the behavior distribution using an-greedy strategy…With this strategy, a random action is taken with probability ε and the greedy action… is chosen with probability 1 − ε” where the selection greedy exploitation(1- ε) corresponds to a first probability and the behavior distribution corresponds to traversing/drawing edges as the states and actions/transitions as described in Baker form a graph(See also Baker, Page 3, Paragraph 7, “We model the layer selection process as a Markov Decision Process with the assumption that a well-performing layer”)) and edges are drawn with an identical probability based on a second probability(Baker, Page 5, Paragraph 4, “At ε = 1.0, the agent samples CNN architecture with a random walk along a uniformly weighted Markov chain” where a random along a uniformly weighted Markov chain corresponds to edges are drawn with an identical probability based on a second probability as a uniformly weighted Markov chain transitions out of a given state choose the next edge with equal probability and the likelihood of edges being drawn with an identical probability being dictated by probability ε corresponds being based on a second probability), wherein a sum of the first probability and the second probability is equal to one(Baker, Page 3, Paragraph 6, “We choose the behavior distribution using an-greedy strategy…With this strategy, a random action is taken with probability ε and the greedy action… is chosen with probability 1 − ε” where the probability of a random exploration(ε) and greedy exploitation(1- ε) summing to 1 corresponds with wherein a sum of the first probability and the second probability is equal to one probability)
References Xie and Baker are analogous art because they are from the same field of endeavor of automated neural network architecture design
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Xie and Baker before him or her, to modify SNAS’s exploratory stochastic edge sampling of Xie to include the epsilon greedy exploration/exploitation of Baker as both are mechanisms to control stochastic exploration and the epsilon greedy method has the capacity to handle large or continuous state spaces. The suggestion/motivation for doing so would have been “The number of possible choices makes the design space of CNN architectures extremely large and hence, infeasible for an exhaustive manual search”(Baker, Page 1, Paragraph 1) and “However, in the case of large or continuous state spaces, the-greedy strategy of learning has been empirically shown to converge” ”(Baker, Page 2, Paragraph 4)
Claims 2 are rejected under 35 U.S.C. 103 as being obvious over Xie et al. (SNAS: STOCHASTIC NEURAL ARCHITECTURE SEARCH, henceforth known as Xie) in view of Baker et al. (DESIGNING NEURAL NETWORK ARCHITECTURES USING REINFORCEMENT LEARNING, henceforth known as Baker) and further in Chen et al. (DRNAS: DIRICHLET NEURAL ARCHITECTURE SEARCH, henceforth known as Chen)
Regarding claim 2:
The rejection of claim 1 with prior art Xie-Baker is incorporated and further:
Chen discloses wherein, when a measure of the distribution of the values of the respective variables(Chen, “we select Dirichlet distribution to model its behavior, i.e., q(θ|β) ~ Dir(β), where β represents the Dirichlet concentration parameter” where β is the Dirichlet concentration distribution learned parameter controlling sampling and β is considered a measure of a distribution) relative to a predefined target measure of a target distribution(Chen, Page 3, Paragraph 5, “Therefore, we add a penalty term in the objective (2) to regularize the distance between and the anchor β* = 1, which corresponds to a symmetric Dirichlet” where β* is considered a predefined target measure of a target distribution) is greater(Chen, Page 4, Equation 5 and Proposition 1, where the operationally if β >1 the penalty term is increased to bring β closer to 1) , the respective variables are changed in such a way that edges having an essentially equal probability are drawn(Chen, Page 3, Paragraph 5, “Therefore, we add a penalty term in the objective (2) to regularize the distance between and the anchor β* = 1, which corresponds to a symmetric Dirichlet” where a symmetric Dirichlet is a uniform distribution and is considered an equal probability as all concentration parameters equal to 1)
References Xie and Chen are analogous art because they are from the same field of endeavor of automated design and optimization of neural architecture search(NAS) for improved performance of machine learning tasks.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Xie and Chen before him or her, to modify the Gumbel-SoftMax of Xie to include the Dirichlet sampling of Chen to model probabilities on each edge as a distribution and maintain more control over exploration, exploitation and sparse sampling. The suggestion/motivation for doing so would have been “The concentration parameter β controls the sampling behavior of Dirichlet distribution and is crucial in balancing exploration and exploitation during the search phase. Let βo denote the concentration parameter assign to operation o. When βo << 1 for most o = 1 ~ |O|, Dirichlet tends to produce sparse samples with high variance, reducing the training stability; when βo >> 1 for most o = 1 ~ |O|, the samples will be dense with low variance, leading to insufficient exploration.”(Chen, Page 3, Paragraph 5)
Claim(s) 3-4 is/are rejected under 35 U.S.C. 103 as being unpatentable over Xie et al. (SNAS: STOCHASTIC NEURAL ARCHITECTURE SEARCH, henceforth known as Xie) Baker et al. (DESIGNING NEURAL NETWORK ARCHITECTURES USING REINFORCEMENT LEARNING, henceforth known as Baker) and further in view of Haarnoja et al. (Soft Actor-Critic Algorithms and Applications, henceforth known as Haarnoja)
Regarding claim 3:
The rejection of claim 1 with prior art Xie-Baker is incorporated and further:
Haarnoja discloses wherein the change of the respective variables takes place as a function of an entropy of the directed graph(Haarnoja, Page 7, Paragraph 2, “Our aim is to find a stochastic policy with maximal expected return that satisfies a minimum expected entropy constraint” where minimum expected entropy constraint is considered a predefined target entropy), and a number of training steps which have already been carried out(Haarnoja, Page 2, Paragraph 2, “To resolve this issue, we devise an automatic gradient-based temperature tuning method that adjusts the expected entropy over the visited states to match a target value” where the temperature(entropy control) adapts continuously as training proceeds is considered changing respective variables as a function of entropy and a number of training steps)
References Xie and Haarnoja are analogous art because they are from the same field of endeavor of using gradient-based optimization of training deep networks.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Xie and Haarnoja before him or her, to modify the Gumbel-SoftMax of Xie to include the entropy evaluation of Haarnoja to maximize randomness while achieving a goal to encourage trying various actions instead of narrowing in on a deterministic strategy. The suggestion/motivation for doing so would have been “The maximum entropy objective has a number of conceptual and practical advantages. First, the policy is incentivized to explore more widely, while giving up on clearly unpromising avenues. Second, the policy can capture multiple modes of near-optimal behavior. In problem settings where multiple actions seem equally attractive, the policy will commit equal probability mass to those actions”(Haarnoja, Page 4, Paragraph 3)
Regarding claim 4:
The rejection of claim 3 with prior art Xie-Baker-Haarnoja is incorporated and further:
Haarnoja discloses when the entropy is greater than a predefined target entropy(Haarnoja, Page 12, Equation 18, where the H is the predefined target entropy and -logπt(at|st) is the instantaneous/current entropy for the sampled action at at state st and the gradient finds the difference between the predefined target and instantaneous entropy), a parameter by which the respective variables are changed is changed in such a way that it changes values of the respective variables (Haarnoja, Page 12, Equation 18, when current entropy is greater than target entropy, the gradient is negative and the temperature(α) decreases which decreases the emphasis on the entropy term in policy objective, which is considered changing respective variables as this will change training/sampling by altering exploration behavior and learning trajectory), so that the probability distribution characterizing the respective variables has a lesser similarity to a uniform distribution(Haarnoja, Page 12, Equation 18, where a lower temperature produces a sharper probability distribution and is considered less similar to a uniform distribution)
Haarnoja discloses and when the ascertained entropy is smaller than the predefined target entropy(Haarnoja, Page 12, Equation 18, where the H is the predefined target entropy and -logπt(at|st) is the instantaneous/current entropy for the sampled action at at state st and the gradient finds the difference between the predefined target and instantaneous entropy), the parameter is changed in such a way that it changes values of the respective variables(Haarnoja, Page 12, Equation 18, when current entropy is less than target entropy, the gradient is positive and the temperature(α) increases which increases the emphasis on the entropy term in policy objective, which is considered changing respective variables as this will change training/sampling by altering exploration behavior and learning trajectory), so that the probability distribution characterizing the respective variables characterizes a uniform distribution(Haarnoja, Page 12, Equation 18, where a higher temperature produces a flatter probability distribution and is considered more similar to a uniform distribution)
Response to Arguments
Applicant's arguments filed 04/22/2026 have been fully considered but they are not persuasive. A breakdown can be found below:
101:
Applicant appears to argue on page 6-7 that the limitation training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs, wherein during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized recites significantly more and overcomes 101.
Examiner respectfully disagrees as training is not described in such a way to realize the cost optimization cited in the arguments as the current claims recite the well understood, routine and conventional activity of training by adjusting variables/parameters to optimize a function.
Applicant appears to argue on page 7-10 that a technological improvement citing to the specification and the use of an epsilon-greedy exploration probability for decision making for subgraph/edge selection generation. Specifically, Applicant asserts the use of epsilon-greedy exploration increases the likelihood that the search will result in conservation of computation resources and produce higher-equality solutions which provides a direct improvement in a computer’s performance. Applicant cites to Ex parte Desjardins and cites the limitation training a machine learning system corresponding to a drawn subgraph of the multitude of subgraphs, wherein during the training, parameters of the machine learning system and the respective variables are adapted so that a cost function is optimized as the limitation that reflects the improvements described because “a subgraph of a multitude of subgraphs” has respective variables that are changed as a function of the first probability which improves neural architecture search and premature fixation.
Examiner respectfully disagrees as Examiner’s review of the arguments understands the Applicant arguments as an improvement provided by the claimed abstract idea drawing edges using mathematical epsilon-greedy algorithm, which does not result in an improvement in technology. Applicants example describes an improvement provided by the claimed abstract idea of drawing edges using mathematical epsilon-greedy algorithm which does not result in a improvement in technology as an additional element does not provide or reflect the improvement.
Additionally, although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). Applicant appears to be interpreting a narrower claim as the current claims do not positively recite additional elements that provide a technological improvement.
Further, claims do not reflect the desired outcomes of increasing the likelihood that the search will result in conservation of computation resources and produce higher-equality solutions. Applicants arguments only states an improvement of desired outcomes however no specific steps or limitations of the training or graphs that reflects the optimization.
102/103:
Applicant appears to argue on page 10-12 that Xie does not discloses the amended language of drawing a multitude of subgraphs as a function of respective variables that change as a function of a first probability, such that edges are drawn as a function of the respective variables assigned to the edges based on the first probability
Examiner respectfully disagrees as Xie explains Zi,j is a the variable sampled from the distribution p(Z) that determines which operation O is active on the edge i,j between node i and j(Xie, Page 2, Figure 1, “Sampled from p(Z), Z is a matrix whose rows Zi,j are one-hot random variable vectors…Columns of this matrix correspond to operations Ok”)
Additionally, this Zi,j is used to obtain a child graph by determining which edge is to be drawn(Xie, page 3, Paragraph 4, “Multiplying each one-hot random variable Zi,j to each edge (i,j) in the DAG, we obtain a child graph”)
Applicant appears to argue on page 10-12 that Xie does not discloses the amended language of drawing a multitude of subgraphs…as a function of a first probability, such that edges are drawn as a function of the respective variables assigned to the edges based on the first probability and such that edges are drawn with an identical probability based on a second probability, with the first and second probabilities summing to one.
Applicant’s arguments with respect to the cited limitations have been considered but are moot because the new ground of rejection relies on new art, Baker, in combination of Xie to teach the amended limitation.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHARLES JEFFREY JONES JR whose telephone number is (703)756-1414. The examiner can normally be reached Monday - Friday 8:00 - 5:00 EST.
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, Kakali Chaki can be reached at 571-272-3719. 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.
/C.J.J./Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights