DETAILED ACTION
1. This communication is in response to the amendments filed December 15, 2025 for Application No. 17/936,254 in which Claims 1-20 are presented for examination.
Notice of Pre-AIA or AIA Status
2. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
3. The arguments/remarks filed on December 15, 2025 have been considered. Claims 1-7 and 9-20 have been amended. Thus, Claims 1-20 are pending and presented for examination.
4. Applicant’s arguments filed December 15, 2025 with respect to the 35 U.S.C. 112(b) rejection and 35 U.S.C. 112(f) interpretation have been fully considered and are persuasive. Thus, the 35 U.S.C. 112(b) rejection and 35 U.S.C. 112(f) interpretation have been withdrawn. Examiner Note: It should be noted that a new 35 U.S.C. 112(b) rejection is issued below, as necessitated by amendment.
5. Applicant's arguments filed December 15, 2025 with respect to the 35 U.S.C. 101 rejection have been fully considered but they are not persuasive.
Applicant’s Arguments on Pg. 8 of Arguments/Remarks state:
“Claim 1 recites a device that generates a "degenerated network data structure... by controlled modification of an adjacency matrix." An "adjacency matrix" is a specific computer data structure used to represent graph topology in memory; it is not an abstract idea or a mental concept. The claim requires the physical manipulation of this matrix to satisfy a parameter set.
While the claims involve mathematical parameters (e.g., complex network parameters), they use these parameters to define the "macroscopic characteristics" of the data structure. The claim does not recite the calculation of parameters in isolation; it recites a tool that uses the parameters to constrain the modification of a specific data object.
This mirrors USPTO Example 39, where the claims "involved" math to train a neural network but were eligible because they were directed to the technological process of training. Here, the claims are directed to the technological process of generating structurally reliable training data (the "degenerated network data structure").”
Examiner respectfully disagrees. First, an “adjacency matrix” is not exclusively a computer data structure. Adjacency matrices are utilized in mathematical concepts/processes, especially related to graphs & graph theory – while these matrices may be used as a data structure for the representation of graphs in computer programs, this does not limit adjacency matrices to only being used as a representation in computer memory. At Step 2A Prong 1, the limitation “generate, by controlled modification of an adjacency matrix of the original data structure, a degenerated network data structure […]” may still be practically performed by mental and/or mathematical process. For example, a user may perform controlled modification of an adjacency matrix of the original network data structure (i.e., with the aid of pen and paper, directly adding/removing edges, reordering rows and columns, resizing the matrix, replacing weights, etc. in order to reduce the influence of noisy links and compensate for missing or unobserved links) to accordingly generate a degenerated network data structure within a predetermined error range, which differs from the original data structure. Furthermore, this controlled modification of the adjacency matrix may also be performed by mathematical process involving matrix addition, scalar multiplication, and structural operations to update edge relationships. An adjacency matrix amounts to no more than a simple tabular representation of edges/weights in a standard weighted graph – a human user is capable of modifying such an adjacency matrix by merely observing/analyzing a graph structure and accordingly using judgement/evaluation to generate and/or modify a matrix representing the graph structure’s edges, links, and weights.
Regarding Applicant’s argument that the instant claims mirror example 39, Examiner respectfully disagrees. Example 39 outlines specific steps which precede the technological process of training and may not be practically performed by mental process. In comparison, the instant claims merely recite constructing a network structure, which may be interpreted as a graph structure, extracting parameters defining macroscopic characteristics (such as a number of nodes, links, etc.), and then generating a degenerated network data structure through controlled modification of an adjacency matrix. These steps may all be feasibly performed by mental process. For further explanation and examples, please see the subsequent 35 U.S.C. 101 section below.
Applicant’s Arguments on Pg. 9 of Arguments/Remarks state:
“Even if the claims were viewed as reciting an abstract idea, the amended limitations integrate the exception into a practical application that improves computer functionality, consistent with Ex parte Desjardins (PTAB 2025).
The amended claims require generating a "degenerated network data structure" by "controlled modification of an adjacency matrix" to "reduce the influence of noise links and to compensate for missing or unobserved links."
As noted in the Specification, conventional graph data often contains "noise links" or "missing links" that degrade AI model performance [Spec. 0005-0006]. Simple augmentation techniques like random dropout "do not sufficiently consider context" [Spec. 0003]. By generating an ensemble of graph data structures (Amended Claim 1) that share macroscopic parameters but vary in microscopic connectivity (the modified adjacency matrix), the claimed invention allows the system to "remove noise from a statistical point of view" [Spec. 0136].
This is a specific improvement to the structural reliability of the data used for computer- based network processing. It transforms a generic computer into a specialized apparatus for robust data augmentation, which Desjardins confirms is a patent-eligible improvement to the functioning of the computer itself.
The Office argues the claims are "mental processes." However, claim 1 requires receiving data from an "external data source" (e.g., WWW, protein interactions [Spec. 0047]) and using "one or more processors" to perform a "controlled modification of an adjacency matrix." Calculating spectral properties (eigenvalues) or performing Monte-Carlo link exchanges on an adjacency matrix for a graph with thousands or millions of nodes (like the WWW) is computationally impossible for a human to perform mentally or with pen and paper. The Examiner has provided no evidence to support the assertion that such complex operations on high-dimensional graph data structures are mental processes. Therefore, the rejection fails to meet the preponderance of evidence standard required by the August 2025 Memo.
The amended claims are directed to a device that generates specific graph data structures to improve the structural reliability of data for AI systems. They do not recite abstract ideas in isolation, and they provide a concrete technical solution to the problem of noise in graph data. Applicant requests withdrawal of the Section 101 rejection.”
Examiner respectfully disagrees. The “controlled modification of an adjacency matrix” to “reduce the influence of noise links and to compensate for missing or unobserved links” may still be practically performed by mental process, as stated above. For example, a user may modify the adjacency matrix (by directly adding/removing edges, reordering rows and columns, resizing the matrix, replacing weights, etc.) in order to reduce the influence of noisy links and compensate for missing or unobserved links – the user may observe/analyze the structure and accordingly identify such noisy/missing/unobserved links and use judgement/evaluation to reduce the influence of these links.
Furthermore, the claims do recite mental processes, as outlined above. Applicant argues that claim 1 requires receiving data from an “external data source” and provides examples of “www” and “protein interactions” – however, this is not what is recited in the claims. The claims simply state that the data is received from an “external data source” without significantly more – thus, they are interpreted broadly. Hence, considering this interpretation, it is feasible for a human to perform the aforementioned operations on a simple graph structure (not necessarily a graph structure comprising thousands or millions of nodes, as this is not indicated by the claim language). Further, the claims still recite a plurality of mathematical processes, such as calculating spectral properties or performing Monte-Carlo link exchanges.
Thus, the 35 U.S.C. 101 rejection is maintained.
6. Applicant’s arguments filed December 15, 2025 with respect to the 35 U.S.C. 102/103 rejections have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Objections
7. Claim 10 is objected to because of the following informalities:
Claim 10 recites a typographical error “[…] alternative network dta structures […]” and should be corrected to instead recite “[…] alternative network data structures […]”.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
8. The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
9. Claims 1, 14, and their respective dependents 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. Claims 1, 14, and their respective dependents recite the limitation “[…] performed to reduce the influence of noise links and to compensate for missing or unobserved links […]”. However, it is not clear nor defined by the claim, how the controlled modification of the adjacency matrix may “compensate for missing or unobserved links” without significantly more (i.e., are the missing/unobserved links compensated for by adjusting edge weights, are the missing/unobserved links compensated for by value imputation, are the missing/unobserved links compensated for by removing these values from processing completely, etc.) Moreover, Applicant’s specification does not provide further information regarding “compensating” for these missing/unobserved links. This renders the claims indefinite.
Claim Rejections - 35 USC § 101
10. 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.
11. 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: Claim 1 is a system type claim. Therefore, Claims 1-10 are directed to either a process, machine, manufacture, or composition of matter.
2A Prong 1: If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation by mathematical calculation but for the recitation of generic computer components, then it falls within the “Mathematical Concepts” grouping of abstract ideas.
[…] generating a network with improved structural reliability for computer based network processing […] (mental process – generating a network with improved structural reliability may be performed manually by a user observing/analyzing various features and parameters of the network and accordingly using judgement/evaluation to generate a network with features and parameters included which improve structural reliability, based on said analysis)
construct an original network structure representing relationships between nodes based on data received from an external data source (mental process – constructing an original network structure may be performed manually by a user observing/analyzing the received data and accordingly using judgement/evaluation to construct an original network (with aid of pen and paper) representing relationships between nodes, based on said data. For example, a user is able to observe/analyze data received from an external data source and use judgement/evaluation to form a graph, comprising nodes and links, (with aid of pen and paper) based on said analysis)
extract from the original network data structure a parameter set including complex network parameters that define macroscopic characteristics of the original network data structure (mental process – extracting a parameter set may be performed manually by a user observing/analyzing the original network data structure and its respective parameters and accordingly using judgement/evaluation to extract a parameter set (with aid of pen and paper) consisting of complex parameters that define macroscopic characteristics (i.e., number of nodes, number of links, etc. as supported by instant dependent claim 5) from said original network data structure. For example, a user may observe/analyze the original network structure and use judgement/evaluation to determine the number of nodes and number of links, hence extracting a parameter set including complex network parameters)
generate, by controlled modification of an adjacency matrix of the original network data structure, a degenerated network data structure that satisfies the parameter set within a predetermined error range and that differs from the original network data structure in microscopic connectivity, wherein the controlled modification is performed to reduce the influence of noise links and to compensate for missing or unobserved links in the original network data structure (mental/mathematical process – generating a degenerated network data structure that satisfies the parameter set within a predetermined error range may be performed manually by a user performing controlled modification of an adjacency matrix of the original network data structure (i.e., with the aid of pen and paper, directly adding/removing edges, reordering rows and columns, resizing the matrix, replacing weights, etc. in order to reduce the influence of noisy links and compensate for missing or unobserved links) to accordingly generate a degenerated network data structure within a predetermined error range, which differs from the original data structure. Furthermore, this controlled modification of the adjacency matrix may also be performed by mathematical process involving matrix addition, scalar multiplication, and structural operations to update edge relationships)
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
a device […] comprising: one or more processors configured to execute instructions to […] (recited at a high-level of generality (i.e., as a generic device comprising generic computer components which are already configured to perform the specific operations of claim 1) such that it amounts to no more than mere instructions to apply the exception using generic computer components)
2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
a device […] comprising: one or more processors configured to execute instructions to […] (mere instructions to apply the exception using generic computer components cannot provide an inventive concept)
For the reasons above, Claim 1 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 2-10. The additional limitations of the dependent claims are addressed below.
Regarding Claim 2:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 2 depends on.
[…] calculate a score for similarity between the original network data structure and the degenerated network data structure and select an alternative network data structure for the original network data structure from the degenerated network data structure based on the score (mental process – calculating a similarity score may be performed manually by a user observing/analyzing the original network data structure and degenerated network data structure and accordingly using judgement/evaluation to compute a score based on the similarity of both network data structures. Correspondingly, a user may use judgement/evaluation to select an alternative network data structure based on said score)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 3:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 3 depends on.
[…] construct the original network data structure in such a way that a network between variables is configured based on at least any one of similarity between the variables constituting the data and an amount of mutual information (mental process –constructing the original network data structure may be performed manually by a user observing/analyzing the similarity score and the amount of mutual information and accordingly using judgement/evaluation to apply this analysis to the construction of the original network data structure (with aid of pen and paper))
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 4:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 4 depends on.
[…] construct the parameter set with non-contradictory parameters among the complex network parameters extracted from the original data structure (mental process –constructing the parameter set may be performed manually by a user observing/analyzing the parameters extracted and accordingly using judgement/evaluation to construct the parameter set with only non-contradictory parameters)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 5:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 5 depends on.
[…] extract, from the original network data structure, a complex network parameter including at least any one of the number of nodes, the number of links, a degree, a degree distribution, assortativity, a degree correlation, a clustering coefficient, an average shortest path length, centrality, a community structure, motif, a network significance profile (SP), global efficiency, local efficiency, and a spectral property of graph Laplacian (mental process – extracting a complex network parameter may be performed manually by a user observing/analyzing the original network data structure and accordingly using judgement/evaluation to extract different parameters (such as number of nodes, number of links, etc.) based on said analysis)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 6:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 6 depends on.
[…] generate the degenerated network data structure that satisfies the parameter set within a predetermined error range but has a different phenotype from the original network data structure (mental process –generating the degenerated network data structure may be performed manually by a user observing/analyzing the parameter set and accordingly using judgement/evaluation to generate a degenerated network data structure with a different phenotype from the original network)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 7:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 7 depends on.
[…] generate the degenerated network data structure through at least any one of a perturbation method and a link exchange method according to a Monte-Carlo process based on the original network data structure, and the perturbation method performs at least any one of node addition, node deletion, link addition, and link deletion (mental/mathematical process –generating the degenerated network data structure may be performed manually by a user and/or by mathematical process using a perturbation method and a link exchange method according to a Monte-Carlo process. For example, a user may perform the perturbation method comprising node addition/deletion or link addition/deletion with the aid of pen and paper)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 8:
Step 2A Prong 1:
See the rejection of Claim 1 above, which Claim 8 depends on.
Step 2A Prong 2 & Step 2B:
the degenerated network generation unit generates the degenerated network using at least any one of a Barabasi-Albert model, an Erdos-Renyi model, a Watts-Strogatz model, a copying model, an edge inheritance model, a Bianconi-Barabasi model, a fitness model, an aging model, a Dorogovshev-Mendez-Samukin model, an initial attractive model, a nonlinear preferential attachment model, and an accelerated growth model (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying the known methods by which to generate the degenerated network does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 9:
Step 2A Prong 1:
See the rejection of Claim 2 above, which Claim 9 depends on.
Step 2A Prong 2 & Step 2B:
[…] calculate the score using at least any one of Shannon entropy, a spectral method, cosine similarity, an inner product, and a Euclidean distance (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying the known methods to calculate the similarity score does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 10:
Step 2A Prong 1: See the rejection of Claim 2 above, which Claim 10 depends on.
[…] select the alternative network data structure according to any one of a method of selecting a predetermined number of alternative network data structures from the degenerated network data structure in an order of a highest score and a method of selecting a degenerated network having a score greater than or equal to a predetermined threshold as an alternative network data structure (mental process – selecting the alternative network data structure may be performed manually by a user observing/analyzing the alternative network data structures and their corresponding scores and accordingly using judgement/evaluation to select a number of alternative network data structures based on having the highest score and/or a score greater than or equal to a predetermined threshold)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 11:
Step 1: Claim 11 is a system type claim. Therefore, Claims 11-13 are directed to either a process, machine, manufacture, or composition of matter.
2A Prong 1: If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation by mathematical calculation but for the recitation of generic computer components, then it falls within the “Mathematical Concepts” grouping of abstract ideas.
construct an original network data structure representing relationships between nodes for data received from an external data source (mental process – constructing an original network structure may be performed manually by a user observing/analyzing the received data and accordingly using judgement/evaluation to construct an original network (with aid of pen and paper) representing relationships between nodes, based on said data. For example, a user is able to observe/analyze data received from an external data source and use judgement/evaluation to form a graph, comprising nodes and links, (with aid of pen and paper) based on said analysis)
generate an alternative network data structure in which at least one parameter is identical to a parameter extracted from the original network data structure (mental process – generating an alternative network data structure may be performed manually by a user observing/analyzing the parameters extracted from the original network and accordingly using judgement/evaluation to generate an alternative network data structure in which at least one parameter is identical to an extracted parameter from the original network, based on said analysis)
generate a network set composed of the original network data structure and the alternative network data structure (mental process – generating a network set may be performed manually by a user observing/analyzing both networks and accordingly generating a network set composed of both the original network and the alternative network)
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
a network self-supervised learning device comprising: one or more processors configured to execute instructions to […] (recited at a high-level of generality (i.e., as a generic network self-supervised learning device comprising one or more processors without significantly more) such that it amounts to no more than mere instructions to apply the exception using generic computer components)
[…] train a network encoder using a network sampled from the network set […] (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model with previously determined data)
wherein the encoder receives a network data structure […] (Adding insignificant extra-solution activity to the judicial exception – see MPEP 2106.05(g))
[…] transforms the received network data structure into a representation in a latent space. (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of applying an already trained encoder neural network to perform the specific operations of the claim without significantly more)
2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
a network self-supervised learning device comprising: one or more processors configured to execute instructions to […] (mere instructions to apply the exception using generic computer components cannot provide an inventive concept)
[…] train a network encoder using a network sampled from the network set […] (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model with previously determined data)
wherein the encoder receives a network data structure […] (MPEP 2106.05(d)(II) indicates that merely “Receiving or transmitting data over a network” is a well-understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed limitation is well-understood, routine, conventional activity is supported under Berkheimer)
[…] transforms the received network data structure into a representation in a latent space. (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of applying an already trained encoder neural network to perform the specific operations of the claim without significantly more)
For the reasons above, Claim 11 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 12-13. The additional limitations of the dependent claims are addressed below.
Regarding Claim 12:
Step 2A Prong 1: See the rejection of Claim 11 above, which Claim 12 depends on.
sample network pairs from the network set (mental process – sampling network pairs from the network set may be performed manually by a user observing/analyzing the network set and accordingly using judgement/evaluation to sample according network pairs)
[…] generate a representation of the network pairs (mental process – generating a representation of the network pairs may be performed manually by a user observing/analyzing the network pairs and accordingly using judgement/evaluation to generate an according “representation” based on said analysis)
[…] generate a projection for the representation (mental process – generating a projection for the representation may be performed manually by a user observing/analyzing the representation and accordingly using judgement/evaluation to generate an according “projection” based on said analysis)
[…] calculate a mutual information amount between the network pairs based on the projection, and calculate a loss of similarity between the network pairs based on the mutual information amount […] (mental/mathematical process – calculating a mutual information amount and a loss of similarity may be performed by manually by a user observing/analyzing the network pairs and accordingly using judgement/evaluation to calculate a mutual information amount and loss of similarity and/or by mathematical process leveraging formulas/algorithms to calculate mutual information amount and loss of similarity (see dependent claim 13 for formulas/algorithms))
Step 2A Prong 2 & Step 2B:
[…] input the network pairs to the encoder […] (MPEP 2106.05(d)(II) indicates that merely “Receiving or transmitting data over a network” is a well-understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim. Thereby, a conclusion that the claimed limitation is well-understood, routine, conventional activity is supported under Berkheimer)
[…] train the encoder based on the loss (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 11.
Regarding Claim 13:
Step 2A Prong 1:
See the rejection of Claim 12 above, which Claim 13 depends on.
Step 2A Prong 2 & Step 2B:
[…] calculate the loss using at least any one of Kullback-Leibler divergence and an information noise-contrastive estimator (InfoNCE) (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying the formulas used to calculate loss does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 11.
Independent Claim 14 recites substantially the same limitations as Claim 1, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
For the reasons above, Claim 14 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 15-20. The additional limitations of the dependent claims are addressed below.
Claim 15 recites substantially the same limitations as Claim 2, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 16 recites substantially the same limitations as Claim 3, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 17 recites substantially the same limitations as Claim 4, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 18 recites substantially the same limitations as Claim 6, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 19 recites substantially the same limitations as Claim 7, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim 20 recites substantially the same limitations as Claim 9, in the form of a method. The claim is also directed to performing mental processes/mathematical calculations without significantly more, therefore it is rejected under the same rationale.
Claim Rejections - 35 USC § 103
12. 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.
13. Claims 1-10 and 14-20 are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (hereinafter Lee) (“Reconstructing Damaged Complex Networks Based on Neural Networks”), in view of Tacchi et al. (hereinafter Tacchi) (U.S. Patent 9558265).
Regarding Claim 1, Lee teaches a device for generating a network with improved structural reliability for computer based network processing (Lee, Pg. 1, “In this paper, we formulate the network reconstruction problem as an identification of network structure based on much reduced link information. Furthermore, a novel method based on multilayer perceptron neural network is proposed as a solution to the problem of network reconstruction. Based on simulation results, it was demonstrated that the proposed scheme achieves very high reconstruction accuracy in small-world network model and a robust performance in scale-free network model.”, therefore, methods for generating a reconstructed network with improved structural reliability is disclosed), the device comprising: one or more processors configured to execute instructions (While Lee discloses a method based on a multilayer perceptron neural network, implemented on a computer, Lee does not explicitly disclose a device for generating a network with improved structural reliability for computer based network processing, comprising one or more processors – See introduction of Tacchi reference below for teaching of a device comprising one or more processors) to:
construct an original network structure representing relationships between nodes based on data received from an external data source (Lee, Pg. 5, Algorithm 1, “2: Generate a set of M complex networks. 3: Set number of hidden layers. 4: Set number of neurons in each hidden layers. 5: Define activation function in hidden layers. 6: Set number of training iteration Iteration Num”, thus, during the initialization process shown in Algorithm 1, a number of complex networks may be constructed, representing relationships between nodes (See subsequent section 4.1 on Pg. 6 which describes such links between nodes) based on data received externally);
extract from the original network data structure a parameter set including complex network parameters that define macroscopic characteristics of the original network data structure (Lee, Pg. 6, “The MLPNN used in our method has two hidden layers with 64 neurons in the first layer and four neurons in the second layer. The nonlinear activation function in the hidden layer is chosen to be triangular activation function. The number of inputs to the MLPNN depends on the number of nodes in the network. To train and 6 of 11 test the MLPNN, using complex networks with N =10, N =30, and N =50, the number of inputs are set equal to the possible number of node pair combinations, which are 45, 435, and 1225, respectively. As for the number of outputs, eight are chosen to represent maximum number of 256 complex networks. The training input and output data patterns are randomly chosen from LL of M damaged networks with different percentage f of failed nodes out of total N nodes and corresponding indices of the complex networks”, therefore, a parameter set including complex network parameters (such as the number of nodes and/or node degree, supported by preceding section 3.2 and subsequent sections 4.2 and 4.3) that define macroscopic characteristics of the original network data structure are extracted); and
generate, by controlled modification of an adjacency matrix of the original network data structure, a degenerated network data structure (Lee, Pg. 5, “Obtain adjacency matrix for the damaged complex networks m. 13: Transform adjacency matrix into LL.”, therefore, degenerated network data structures may be generated by controlled modification of an adjacency matrix of the original network structure) that satisfies the parameter set within a predetermined error range and that differs from the original network data structure in microscopic connectivity (Lee, Pg. 6, “To evaluate the performance of the proposed method in small-world network model, the network is implemented based on the algorithm described in Figure 1. Figure 6 and Table 1 shows the 4.2. Small-World Network reconstruction error probability as a function of percentage of random node failure f. Furthermore, we study the influence of the number of node on the network reconstruction performance with N = 10, N =30, and N = 50. The initial degree K of the network is set to two and the links are randomly To evaluate the performance of the proposed method in small-world network model, the network is implemented based on the algorithm described in Figure 1. Figure 6 and Table 1 shows the reconstruction error probability p = 0.15”, thus, the small-world network model may be implemented based on Lee Algorithm 1 & thus, the degenerated network structure generated may satisfy the parameter set within a predetermined error range (see Figure 6 and Table 1) and differs from the original network data structure in microscopic connectivity (different number/orientation of nodes/links)),
wherein the controlled modification is performed to reduce the influence of noise links and to compensate for missing or unobserved links in the original network data structure (Lee, Pg. 3, “We simulate the damage process on complex networks by considering the random attack model. In the random attack model, nodes are randomly selected and removed. Note that when a node is removed, all the links connected to that node are also removed [25]. To evaluate the performance of the proposed reconstruction method, the difference in the number of links between the original network and the reconstructed network is used and represented as probability of reconstruction error, PRE, that is defined as […]”, therefore, the controlled modification relates to optimizing a reconstruction error, which aims to reduce the influence of noisy links and to compensate for missing links in the original network data structure).
While Lee teaches methods for generating a network with improved structural reliability and implementing a multilayer perceptron neural network (as disclosed above), Lee does not explicitly disclose a device for generating a network with improved structural reliability for computer based network processing, the device comprising: one or more processors configured to execute instructions to: […]
However, Tacchi teaches a device for generating a network with improved structural reliability for computer based network processing (Tacchi, Col. 14 lines 37-42, “In some embodiments, if (e.g., in response to determining that) a score (e.g., related to a shared edge) is determined to not satisfy a threshold score for maintaining a shared edge, the edge may be removed. In some cases, removal of an edge may include designating the edge as removed (e.g., by setting its weight or other value to indicate the removal)”, therefore, a network/graph may be generated with improved structural reliability (unnecessary/weak edges removed) for computer based network processing), the device comprising: one or more processors configured to execute instructions (Tacchi, Col. 26 lines 48-52, “Computing system 1000 may include one or more processors (e.g., processors 1010a-1010n) coupled to system memory 1020, an input/output I/O device interface 1030, and a network interface 1040 via an input/output (I/O) interface 1050.”, thus, a device (computing system 1000 shown in Figure 10) comprising one or more processors (1010a-1010n in Figure 10) configured to execute instructions is disclosed) to: […]
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the methods for generating a network with improved structural reliability, as disclosed by Lee to include wherein the methods are implemented as a device for generating a network with improved structural reliability for computer based network processing, the device comprising one or more processors configured to execute instructions, as disclosed by Tacchi. One of ordinary skill in the art would have been motivated to make this modification to enable the use of a computing device comprising one or more processors which may provide useful inferences/insights and capture important relationships between nodes, that may be used to generate an improved network (Tacchi, Col. 1 lines 38-52, “Through the sophisticated use of computers, inferences that would otherwise be impractical are potentially attainable, even on relatively large collections of documents. In some cases, a graph may represent relationships between objects indicated in (e.g., named entities mentioned in) a collection of documents (e.g., one or more corpora). Objects may be text or referents of the text, e.g., named entities. The nodes of the graph may represent the objects, and the edges may represent the relationships between objects. The relationships may be determined based on the frequency of terms in text describing the respective objects, where the number of edges linking such graph nodes, the edge weights, and distribution of such edges are based on the frequency of the terms in the plain text.”)
Regarding Claim 2, Lee in view of Tacchi teaches the device of claim 1, wherein the one or more processors are further configured to calculate a score for similarity between the original network data structure and the degenerated network data structure and select an alternative network data structure for the original network data structure from the degenerated network data structure based on the score (Tacchi, Col. 1 lines 18-23 & lines 32-42, “A method comprising: obtaining a graph comprising more than 1000 nodes and more than 2000 edges, each of the edges linking two of the nodes and having a value indicating an amount of similarity between objects corresponding to the two linked nodes, and the graph being generated from a natural language processing of a corpus of unstructured documents; after obtaining the graph, selecting a parameter for influencing the graph […] determining a score related to the edge shared with the evaluation node, the score determined based on the value indicating the amount of similarity and a value of the selected influencing parameter for the evaluation node; determining whether the score satisfies a threshold score for maintaining a shared edge; and removing the edge shared with the evaluation node in response to the score not satisfying the threshold score; and preparing, based on the graph resulting from the assessment of the number of edges of the evaluation nodes, instructions to display at least part of the resulting graph”, therefore, a similarity score may be calculated between the original network/graph and the degenerated network (based on controlled modification of the adjacency matrix based on the influencing parameter – See Tacchi Col. 11 lines 9-27) and an alternative network data structure may be selected based on the score. Examiner also notes that Lee Pg. 5 discloses numerous networks being trained as alternative networks, however, Lee does not explicitly disclose a similarity metric)
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the device of claim 1, as disclosed by Lee in view of Tacchi to include wherein the one or more processors are further configured to calculate a score for similarity between the original network data structure and the degenerated network data structure and select an alternative network data structure for the original network data structure from the degenerated network data structure based on the score, as disclosed by Tacchi. One of ordinary skill in the art would have been motivated to make this modification to enable the use of a similarity metric, which may indicate similarity between data structures, in order to generate and select a network with improved structural reliability, based on the assessment of similarity (Tacchi, Col. 17 lines 38-52, “In some embodiments, the similarity measurement module 846 may be operative to determine similarity between objects (e.g., determining a value indicating similarity between the objects by assessing similarity between vectors corresponding to such objects, or other techniques). In some embodiments, the scoring module 848 may be operative to determine respective scores (e.g., related to nodes, their respective edges, their respective adjacent node or adjacent node candidates, etc.). With respect to steps 112 and 312, for example, the scoring module 848 may determine such a score related to a node based on the value indicating the amount of similarity (between the node and its adjacent node or adjacent node candidate), the value of a selected influencing parameter for the node, etc.”).
Regarding Claim 3, Lee in view of Tacchi teaches the device of claim 1, wherein the one or more processors construct the original network data structure in such a way that a network between variables is configured based on at least any one of similarity between the variables constituting the data and an amount of mutual information (Tacchi, Col. 13 lines 30-42, “As indicated in step 112, a score (related to the edge shared with the node) may be determined based on the value indicating the amount of similarity (between the object corresponding to the node and the object corresponding to the adjacent node) and the value of the selected influencing parameter for the node. In one scenario, edge weights are represented as numbers from 0 to 1, where an edge having a weight of 0 indicates a lack of similarity between the two nodes (or between their respective corresponding objects) that the edge connects, and an edge having a weight of 1 indicates a very high amount of similarity between the two nodes (or between their respective corresponding objects) that the edge connects.”, thus, the one or more processors may construct the original network structure in a way that a network between variables is configured based at least any one of similarity between the variables/edges constituting the data and mutual information).
The reasons of obviousness have been noted in the rejection of Claims 1 & 2 above and applicable herein.
Regarding Claim 4, Lee in view of Tacchi teaches the device of claim 1, wherein the one or more processors constructs the parameter set with non-contradictory parameters among the complex network parameters extracted from the original data structure (Lee, Pg. 6, “The number of inputs to the MLPNN depends on the number of nodes in the network. To train and 6 of 11 test the MLPNN, using complex networks with N =10, N =30, andN =50, the number of inputs are set equal to the possible number of node pair combinations, which are 45, 435, and 1225, respectively. As for the number of outputs, eight are chosen to represent maximum number of 256 complex networks. The training input and output data patterns are randomly chosen from LL of M damaged complex networks with different percentage f of failed nodes out of total N nodes and corresponding indices of the complex networks”, therefore, the parameter set may be constructed with “non-contradictory parameters” among the complex network parameters extracted – using broadest reasonable interpretation of the term “non-contradictory parameters”, the complex networks have a different number of nodes, with the number of inputs set equal to the possible number of node pair combinations, hence they are non-contradictory as there is no overlap between the number of nodes and links of networks tested).
Regarding Claim 5, Lee in view of Tacchi teaches the device of claim 1, wherein the one or more processors extracts, from the original network data structure, a complex network parameter including at least any one of the number of nodes, the number of links, a degree, a degree distribution, assortativity, a degree correlation, a clustering coefficient, an average shortest path length, centrality, a community structure, motif, a network significance profile (SP), global efficiency, local efficiency, and a spectral property of graph Laplacian (Lee, Pg. 6, “Furthermore, we study the influence of the number of node on the network reconstruction performance with N = 10, N =30, and N = 50. The initial degree K of the network is set to two and the links are randomly rewired with probability p = 0.15”, hence, as shown with the exemplary small-world network model, the parameters may include number of nodes, number of links, a degree, etc. This holds true for the scale-free network described in section 4.3 as well).
Regarding Claim 6, Lee in view of Tacchi teaches the device of claim 1, wherein the one or more processors generates the degenerated network data structure that satisfies the parameter set within a predetermined error range but has a different phenotype from the original network data structure (Lee, Pg. 7, Tables 1 & 2 depict how the degenerated network data structures (with a different number of nodes and hence a different phenotype) satisfy the parameter set within a predetermined reconstruction error range).
Regarding Claim 7, Lee in view of Tacchi teaches the device of claim 1, wherein the one or more of processors generate the degenerated network data structure through at least any one of a perturbation method and a link exchange method according to a Monte-Carlo process based on the original network data structure, and the perturbation method performs at least any one of node addition, node deletion, link addition, and link deletion (Lee, Pg. 3, “We simulate the damage process on complex networks by considering the random attack model. In the random attack model, nodes are randomly selected and removed. Note that when a node is removed, all the links connected to that node are also removed […]”, therefore, the degenerated network data structure may be generated through a perturbation method including node and/or link deletion).
Regarding Claim 8, Lee in view of Tacchi teaches the device of claim 1, wherein the degenerated network generation unit generates the degenerated network using at least any one of a Barabasi-Albert model, an Erdos-Renyi model, a Watts-Strogatz model, a copying model, an edge inheritance model, a Bianconi-Barabasi model, a fitness model, an aging model, a Dorogovshev-Mendez-Samukin model, an initial attractive model, a nonlinear preferential attachment model, and an accelerated growth model (Lee, Pg. 2, “We evaluate the performance of the proposed method based on simulations in two classical complex networks (1) small-world network and (2) scale-free networks, generated by Watts and Strogatz model and Barabási and Albert model, respectively.”, thus, the networks may be generated using at least any one of a Barabasi-Albert model and a Watts-Strogatz model).
Regarding Claim 9, Lee in view of Tacchi teaches the device of claim 2, wherein the one or more processors calculate the score using at least any one of Shannon entropy, a spectral method, cosine similarity, an inner product, and a Euclidean distance (Tacchi, Col. 23 lines 20-23, “Similarity of n-grams (and corresponding entities) may be determined based on similarity of resulting vectors in the co-occurrence matrix, e.g., based on cosine similarity.”, therefore, the score may be calculated using a cosine similarity).
The reasons of obviousness have been noted in the rejection of Claims 1 & 2 above and applicable herein.
Regarding Claim 10, Lee in view of Tacchi teaches the device of claim 2, wherein the one or more processors select the alternative network data structure according to any one of a method of selecting a predetermined number of alternative network data structures from the degenerated network data structure in an order of a highest score and a method of selecting a degenerated network having a score greater than or equal to a predetermined threshold as an alternative network data structure (Tacchi, Col. 2 lines 32-42, “determining a score related to the edge shared with the evaluation node, the score determined based on the value indicating the amount of similarity and a value of the selected influencing parameter for the evaluation node; determining whether the score satisfies a threshold score for maintaining a shared edge; and removing the edge shared with the evaluation node in response to the score not satisfying the threshold score; and preparing, based on the graph resulting from the assessment of the number of edges of the evaluation nodes, instructions to display at least part of the resulting graph.”, thus, the alternative network data structure may be selected according to any one of a method of selecting based on having a score greater than or equal to a predetermined threshold).
The reasons of obviousness have been noted in the rejection of Claims 1 & 2 above and applicable herein.
Regarding Claim 14, Lee in view of Tacchi teaches a method of generating a network with improved structural reliability (Lee, Pg. 1, “In this paper, we formulate the network reconstruction problem as an identification of network structure based on much reduced link information. Furthermore, a novel method based on multilayer perceptron neural network is proposed as a solution to the problem of network reconstruction. Based on simulation results, it was demonstrated that the proposed scheme achieves very high reconstruction accuracy in small-world network model and a robust performance in scale-free network model.”, therefore, methods for generating a reconstructed network with improved structural reliability is disclosed) for computer-based network processing (Tacchi, Col. 14 lines 37-42, “In some embodiments, if (e.g., in response to determining that) a score (e.g., related to a shared edge) is determined to not satisfy a threshold score for maintaining a shared edge, the edge may be removed. In some cases, removal of an edge may include designating the edge as removed (e.g., by setting its weight or other value to indicate the removal)”, therefore, a network/graph may be generated with improved structural reliability (unnecessary/weak edges removed) for computer based network processing), the method comprising: […]
The rest of the claim language in Claim 14 recites substantially the same limitations as Claim 1, in the form of a method, therefore it is rejected under the same rationale.
The reasons of obviousness have been noted in the rejection of Claim 1 above and applicable herein.
Claim 15 recites substantially the same limitations as Claim 2 in the form of a method, therefore it is rejected under the same rationale.
Claim 16 recites substantially the same limitations as Claim 3 in the form of a method, therefore it is rejected under the same rationale.
Claim 17 recites substantially the same limitations as Claim 4 in the form of a method, therefore it is rejected under the same rationale.
Claim 18 recites substantially the same limitations as Claim 6 in the form of a method, therefore it is rejected under the same rationale.
Claim 19 recites substantially the same limitations as Claim 7 in the form of a method, therefore it is rejected under the same rationale.
Claim 20 recites substantially the same limitations as Claim 9 in the form of a method, therefore it is rejected under the same rationale.
14. Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (hereinafter Lee) (“Reconstructing Damaged Complex Networks Based on Neural Networks”), in view of Baker et al. (hereinafter Baker) (US PG-PUB 20200143240).
Regarding Claim 11, Lee teaches a network self-supervised learning device comprising: one or more processors (See introduction of Baker reference below for teaching of a network self-supervised learning device comprising one or more processors) configured to execute instructions to:
construct an original network data structure representing relationships between nodes for data received from an external data source (Lee, Pg. 5, Algorithm 1, “2: Generate a set of M complex networks. 3: Set number of hidden layers. 4: Set number of neurons in each hidden layers. 5: Define activation function in hidden layers. 6: Set number of training iteration Iteration Num”, thus, during the initialization process shown in Algorithm 1, a number of complex networks may be constructed, representing relationships between nodes (See subsequent section 4.1 on Pg. 6 which describes such links between nodes) based on data received externally), generate an alternative network data structure in which at least one parameter is identical to a parameter extracted from the original network data structure (Lee, Pg. 6, “To evaluate the performance of the proposed method in small-world network model, the network is implemented based on the algorithm described in Figure 1. Figure 6 and Table 1 shows the 4.2. Small-World Network reconstruction error probability as a function of percentage of random node failure f. Furthermore, we study the influence of the number of node on the network reconstruction performance with N = 10, N =30, and N = 50. The initial degree K of the network is set to two and the links are randomly rewired with probability p = 0.15”, thus, an alternative network data structure may have a different number of nodes, but still have a shared initial degree), and generate a network set composed of the original network data structure and the alternative network data structure (Lee, Pg. 5, “Based on the training input data and desired output data, representing network topology of different complex networks, the goal of the MLPNN is to be able to identify, reconstruct, and produce node pair information of the original network, among numerous networks used to train the neural network”, therefore, a network set composed of the original network data structure and the alternative network data structure may be generated); and
Lee does not explicitly disclose:
a network self-supervised learning device comprising: one or more processors configured to execute instructions […]
train a network encoder using a network sampled from the network set, wherein the encoder receives a network data structure and transforms the received network data structure into a representation in a latent space.
However, Baker teaches:
a network self-supervised learning (Baker, Par. [0154], “Learning coach 101 of FIG. 1B is trained by prior experience on other problems to detect these and other patterns. Learning coach 101 may also be trained to discover new useful patterns based on unsupervised or self-supervised learning. In this self-supervised learning, learning coach 101 verifies the performance enhancing value of a putative pattern and an associated decision by measuring the effect of the decision for instances of detection of the putative pattern on multiple example machine learning problems.”, thus, a self-supervised learning module is disclosed) device comprising: one or more processors configured to execute instructions (Baker, Par. [0225], “In various implementations, one or more of the aforementioned methods and steps thereof can be embodied as instructions stored on a memory of a computer system that is coupled to one or more processor cores such that, when executed by the processor cores, the instructions cause the computer system to perform the described steps.”, therefore, a device comprising one or more processors configured to execute instructions is disclosed) […]
train a network encoder using a network sampled from the network set, wherein the encoder receives a network data structure and transforms the received network data structure into a representation in a latent space (Baker, Par. [0057], “In the illustrated aspect, a first autoencoder 621 comprises an encoder 602 (e.g., a deep neural network) and a decoder 605 (e.g., a deep neural network) and a second autoencoder 631 comprises a cluster classifier 604 as an encoder and a decoder 608. The architecture of the double autoencoder 611 forces the neural network to find the sparse intermediate representation 603 or some other low data-bandwidth representation of the provided input 601.”, therefore, an encoder is trained (See Par. [0071] for recitation of training the encoder) using sample input (associated with a network), wherein the encoder receives the input and transforms the received input into a representation in a latent space, hence transforming the network representation).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the methods of claim 11, as disclosed by Lee to include a network self-supervised learning device comprising: one or more processors configured to execute instructions to train a network encoder using a network sampled from the network set, wherein the encoder receives a network and transforms the received network into a representation in a latent space, as disclosed by Baker. One of ordinary skill in the art would have been motivated to make this modification to enable the discovery of new useful patterns and sparse representations, which may be found through the implementation of self-supervised learning and an encoder (Baker, Par. [0057], “The architecture of the double autoencoder 611 forces the neural network to find the sparse intermediate representation 603 or some other low data-bandwidth representation of the provided input 601” & Par. [0154], “Learning coach 101 of FIG. 1B is trained by prior experience on other problems to detect these and other patterns. Learning coach 101 may also be trained to discover new useful patterns based on unsupervised or self-supervised learning”).
15. Claims 12-13 are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (hereinafter Lee) (“Reconstructing Damaged Complex Networks Based on Neural Networks”), in view of Baker et al. (hereinafter Baker) (US PG-PUB 20200143240), in view of Cliff et al. (hereinafter Cliff) (“Minimising the Kullback-Leibler Divergence for Model Selection in Distributed Nonlinear Systems”).
Regarding Claim 12, Lee in view of Baker teaches the device of claim 11, wherein the one or more processors are further configured to:
sample network pairs from the network set (Baker, Par. [0054], “For purpose of future reference, let N be the network that is the subject of the present discussion, i.e., the network to be made more robust. In an illustrative aspect of step 106 of FIG. 1A, the learning coach 101 separates the training data into two or more disjoint subsets, based on the direction vectors of the gradients of the error cost function with respect to the input nodes”, thus, the data associated with a network may be sampled);
input the network pairs to the encoder to generate a representation of the network pairs (Baker, Par. [0057], “In the illustrated aspect, a first autoencoder 621 comprises an encoder 602 (e.g., a deep neural network) and a decoder 605 (e.g., a deep neural network) and a second autoencoder 631 comprises a cluster classifier 604 as an encoder and a decoder 608. The architecture of the double autoencoder 611 forces the neural network to find the sparse intermediate representation 603 or some other low data-bandwidth representation of the provided input 601.”, therefore, an encoder is trained (See Par. [0071] for recitation of training the encoder) using sample input (associated with a network), wherein the encoder receives the input and transforms the received input into a representation in a latent space, hence transforming the network representation);
generate a projection for the representation (Baker, Par. [0059], “The decoder 608 of the second autoencoder 631 further outputs a copy 609 of the sparse representation 603 provided to the second autoencoder 631.”, thus, a decoding unit for decoding the inputted data and correspondingly generate a “projection” of the representation is disclosed);
calculate a mutual information amount between the network pairs based on the projection, and calculate a loss of similarity between the network pairs based on the mutual information amount (See introduction of Cliff reference below for teaching of calculating a mutual information amount and loss of similarity),
wherein the one or more processors trains the encoder based on the loss (Baker, Par. [0078], “In the illustrated exemplary aspect, the autoencoder 821 is trained to produce the clean input data 808, as close as it can, given the noisy data 801. The autoencoder 821 is also trained with the objective of helping classifier 810 have a low cost of classification errors. It is trained to this objective by continuing the backpropagation done in training classifier 810 back through the nodes representing the estimated clean input data 807 and from there back through the autoencoder 821 network. The backpropagation from the clean input data 808 as a target output and the classifier 810 simultaneously trains the autoencoder 821 according to the two objectives.”, thus, the encoder is trained based on backpropagation of loss).
The reasons of obviousness have been noted in the rejection of Claim 11 above and applicable herein.
Lee in view of Baker does not explicitly disclose calculate a mutual information amount between the network pairs based on the projection, and calculate a loss of similarity between the network pairs based on the mutual information amount,
However, Cliff teaches calculate a mutual information amount between the network pairs based on the projection (Cliff, Pg. 11, “Building on the maximum likelihood score (27), we propose using independence tests to define two new scores of practical value. Here, we draw on the result of de Campos [3], who derived a scoring function for BN structure learning based on conditional mutual information and statistical significance tests, called MIT.”, therefore, a mutual information amount may be calculated), and calculate a loss of similarity between the network pairs based on the mutual information amount (Cliff, Pg. 3, “In this paper, we extend this framework by first showing that KL divergence can be decomposed as information-theoretically useful measures, and then arriving at a similar result but employing non-parametric density estimation techniques to allow for no assumptions about the underlying distributions.”, thus, a loss of similarity may be calculated using a Kullback-Leibler divergence),
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the device of claim 12, as disclosed by Lee in view of Baker to include wherein the device can calculate a mutual information amount between the network pairs based on the projection, and calculate a loss of similarity between the network pairs based on the mutual information amount, as disclosed by Cliff. One of ordinary skill in the art would have been motivated to make this modification to enable the calculation of a mutual information amount and similarity, which may provide insights regarding quality of a network, hence enabling improved network selection (Cliff, Pg. 1, “We are interested in inferring data-driven models of such systems, specifically in the case where each subsystem can be viewed as a nonlinear dynamical system. In this context, the Kullback–Leibler (KL) divergence is commonly used to measure the quality of a statistical model [1–3]. When a model is compared with fully observed data, computing the KL divergence can be straightforward. However, in the case of spatially distributed dynamical systems, where individual subsystems are coupled via latent variables and observed through a filter, the presence of hidden variables renders typical approaches unusable. We derive the KL divergence in such systems as a function of two information-theoretic measures using methods from differential topology”)
Regarding Claim 13, Lee in view of Baker in view of Cliff teaches the device of claim 12, wherein the one or more processors calculate the loss using at least any one of Kullback-Leibler divergence and an information noise-contrastive estimator (InfoNCE) (Cliff, Pg. 1, “We approach the problem by using reconstruction theorems to derive an analytical expression for the KL divergence of a candidate DAG from the observed dataset. Using this result, we present a scoring function based on transfer entropy to be used as a subroutine in a structure learning algorithm.”, therefore, the loss may be calculated using a Kullback-Leibler divergence).
The reasons of obviousness have been noted in the rejection of Claim 12 above and applicable herein.
Conclusion
16. Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
17. Any inquiry concerning this communication or earlier communications from the examiner should be directed to Devika S Maharaj whose telephone number is (571)272-0829. The examiner can normally be reached Monday - Thursday 8:30am - 5:30pm.
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, Alexey Shmatov can be reached at (571)270-3428. 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.
/DEVIKA S MAHARAJ/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123