DETAILED ACTION
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 .
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 19 March 2026 has been entered.
Status of Claims
This Office Action is in response to the communication filed on 19 March 2026.
Claims 1-7 and 12-24 are being considered on the merits.
Claim Rejections - 35 USC § 112
Claims 1-2, 4-6, and 23 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 1-2, 4-6, and 23 recites the limitation "the aggregated node embeddings" in the claims, where there is only “an aggregated node embedding” (singular) claimed prior. There is insufficient antecedent basis for this limitation in the claims.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
Claims 1-7 and 12-24 are rejected under 35 U.S.C. 103 as being unpatentable over Appel, et al. (US 2021/0279279 A1; hereinafter “Appel”) in view of Verma, et. al. (US 2023/0289610 A1; hereinafter “Verma”) and further in view of Hamilton, et. al. (arXiv:1709.05584v3 [cs.SI] 10 Apr 2018; hereinafter, “Hamilton”).
Claim 1, Appel teaches:
A machine learning system for facilitating configuration of machine learning models for different purposes based on multipurpose node embeddings generated via aggregation of multi-dimensional data representations of nodes, the machine learning system comprising: (Appel, para. 0030 and 0032: “The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110. In aspects of the invention, the machine learning algorithms 120 are configured to execute a variety of graph-related tasks on graph embeddings (e.g., an embedded graph-under-development 118 generated by the system 110), including but not limited to node classification (e.g., predicting a type of a given node), link prediction (e.g., predicting whether two nodes are linked), community detection (e.g., identifying densely linked clusters of nodes), and network similarity (e.g., determining how similar are two networks or subnetworks).” “In the simplified example show in FIG. 2, the auto-selected embedding technique 202 embed nodes (v, u) so that distances in the embedded graph-under-development 118A reflect node similarities in the graph-under-development 104A. For a graph-under-development G=(V, E), a node embedding is a mapping f: V→Rd where d is the dimensionality of the embedding space, V represents nodes, and E represents edges. For each node (v∈V) a d dimensional representation is created.”)
one or more processors; and (Appel, para. 0079: “Exemplary computer 802 includes processor cores 804, main memory (“memory”) 810, and input/output component(s) 812, which are in communication via bus 803.”)
one or more non-transitory media having instructions recorded thereon that, when executed by the one or more processors, cause operations comprising: (Verma, para. 0052: “In addition, the server system 102 should be understood to be embodied in at least one computing device in communication with the network 112, which may be specifically configured, via executable instructions, to perform as described herein, and/or embodied in at least one non-transitory computer-readable media.”)
in connection with each node of a plurality of nodes of a graph (Appel, para. 0002: “A graph in this context is made up of vertices (also called nodes or points) that are connected by edges (also called links or lines)”), storing, in a database, an aggregated node embedding (Appel, para. 0038: “In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets. In embodiments of the invention, the repository 502 can be implemented as a relational database that is located in memory (e.g., memory 828 shown in FIG. 8) or any other storage location of the system 110.”) that
(i) is an aggregation of at least first and second node embeddings (a) respectively derived from an unsupervised embedding model and a supervised embedding and (b) corresponding to the same node of the graph (Hamilton, pg. 14, fig. 8: “Figure 8: A, Example of a 4-layer graph, where the same nodes occur in multiple different layers. This multi-layer structure can be exploited to regularize learning at the different layers by requiring that the embeddings for the same node in different layers are similar to each other.” Examiner notes Hamilton at sec. 1.1 teaches both unsupervised and supervised learning for the same node in order to learn embeddings) and (ii) has an aggregated total number of features greater than a total number of features of either the first or second node embedding; (Hamilton, sec. 1.1: “Most of the methods we review will optimize this mapping in an unsupervised manner, making use of only information in A and X, without knowledge of a particular downstream machine learning task. However, we will also discuss some approaches for supervised representation learning, where the models make use of classification or regression labels in order to optimize the embeddings”)
based on the aggregated node embeddings having the aggregated total number of features stored in the database, configuring a first machine learning model corresponding to a first task category to incorporate a first feature subset of the aggregated total number of features without incorporating one or more features of a second feature subset of the aggregated total number of features by, in response to the first machine learning model corresponding to the first task category, using the first feature subset as input parameters for inputs to the first machine learning model; and (Hamilton, sec. 3 and 2.3: “However, unlike the node embedding setting, most subgraph embedding approaches are fully supervised, being used for subgraph classification, where the goal is to predict a label associated with a particular subgraph.” “Adapting the GNN framework to use modern recurrent units also allows Li et al. to leverage node attributes for initialization and to use the output of intermediate embeddings of subgraphs” Examiner notes Hamilton teaches a graph neural network machine learning model on only a subgraph to predict a label i.e. a first label or a first task category where the GNN algorithm nodes accumulate inputs from neighbors)
based on the aggregated node embeddings having the aggregated total number of features stored in the database, configuring a second machine learning model corresponding to a second task category to incorporate the second feature subset without incorporating one or more features of the first feature subset by, in response to the second machine learning model corresponding to the second task category different from the first task category, using the second feature subset as input parameters for inputs to the second machine learning model. (Verma, para. 0003: “The learned representations (i.e., embeddings) may further be used to perform tasks such as link prediction, analysis, and so on.” Examiner notes Hamilton teaches label prediction as a first task category where Verma teaches analysis as an example second task category which Hamilton’s subgraph approach may be applied)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel. Appel teaches matching a graph-under-analysis to a technique for embedding the graph-under-analysis; Verma teaches generating a bipartite graph based on historical transaction data. One of ordinary skill would have been motivated to combine the teachings of Verma into Appel in order to achieve the performance boost offered by node embeddings for downstream tasks (Verma, para. 0033).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified. Appel teaches matching a graph-under-analysis to a technique for embedding the graph-under-analysis; Verma teaches generating a bipartite graph based on historical transaction data; Hamilton teaches methods to embed individual nodes as well as approaches to embed entire (sub)graphs. One of ordinary skill would have been motivated to combine the teachings of Hamilton into Appel in order to use the learned embeddings of subgraphs as feature inputs for downstream machine learning tasks (Hamilton, sec. I).
Claim 2, Appel as modified teaches:
A method comprising: storing, in one or more databases, for each node of a plurality of nodes of a graph, an aggregated node embedding that (Appel, para. 0038: “In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets. In embodiments of the invention, the repository 502 can be implemented as a relational database that is located in memory (e.g., memory 828 shown in FIG. 8) or any other storage location of the system 110.”)
(i) is an aggregation of node embeddings corresponding to the node and respectively derived from different embedding models (Verma, para. 0047: “Furthermore, the present disclosure provides significantly more robust solutions because of handling simultaneous/concurrent processor execution (such as applying one or more neural network models over the same input, simultaneously)” Examiner notes Verma teaches applying different models to same input where Appel teaches an aggregation of embeddings) and (ii) comprises an aggregated set of features with features of a plurality of the derived embeddings; (Appel, para. 0039: “Thus, the embedding technique associated with the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs (i.e., the graph-under-development 104, the extracted features generated by the graph analyzer 112, and the desired task 108) that is over the threshold can be identified or recommended by the embedding analyzer module 114 as an embedding technique for the graph-under-development 104 and the desired task 108.”)
after storing the aggregated node embeddings having the aggregated set of features, (Appel, para. 0038: “ In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets.”) configuring a first machine learning model (Appel, para. 0025: “In some embodiments of the invention, the embedding analyzer module can be implemented as a machine learning model and a set of embedding rules that perform their mapping analyses in series such that the output of the first analysis is fed as an input to the second analysis, thereby improving the overall confidence in the embedding technique selection/recommendation.”) corresponding to a first task category (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.” Examiner notes Hamilton teaches multiple task categories including a first task category) to incorporate a first feature subset of the aggregated set of features without incorporating one or more features of a second feature subset of the aggregated set of features by, (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold… Thus, the embedding technique associated with the known graphs, characteristics of the known graphs, and the known overall tasks that have a level of similarity to the inputs that is over the threshold can be identified or recommended by the embedding analyzer module as an embedding technique for the graph-under-development and the desired task.” Examiner notes that for examination purposes only, “incorporating” is interpreted to mean a formal integration of something i.e. something more integral than “including”; examiner notes that Appel teaches inclusion of subset of features belonging to graphs over a threshold and exclude a subset of features belonging to graphs that do not meet the threshold) in response to the first machine learning model corresponding to the first task category, (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.” Examiner notes Hamilton teaches multiple task categories including a first task category) selecting the first feature subset of the aggregated set of features as input parameters [for the first machine learning model]; and (Verma, para. 0134: “The regular transaction model 548 is configured to calculate the account intelligence score of everyday spending for each cardholder of the plurality of cardholders 124a-124c.”)
After storing the aggregated node embeddings having the aggregated set of features, (Appel, para. 0038: “ In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets.”) configured a second machine learning model corresponding to a second task category to incorporate the second feature subset (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”) without incorporating one or more features of the first feature subset by, (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold… Thus, the embedding technique associated with the known graphs, characteristics of the known graphs, and the known overall tasks that have a level of similarity to the inputs that is over the threshold can be identified or recommended by the embedding analyzer module as an embedding technique for the graph-under-development and the desired task.” Examiner notes that for examination purposes only, “incorporating” is interpreted to mean a formal integration of something i.e. something more integral than “including”; examiner notes that Appel teaches inclusion of subset of features belonging to graphs over a threshold and exclude a subset of features belonging to graphs that do not meet the threshold) in response to the second machine learning model corresponding to the second task category different from the first task category (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.” Examiner notes Hamilton teaches multiple task categories including a first task category) selecting the second feature subset of the aggregated set of features as input parameters for [the second machine learning model]. (Verma, para. 0135: “The discretionary transaction model 550 is configured to calculate the account intelligence score of discretionary spend for each cardholder of the plurality of cardholders 124a-124c.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel as set forth above with respect to claim 1.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claims 3 and 14, Appel as modified teaches:
configuring the first machine learning model comprises selecting the first feature subset of the aggregated set of features as input parameters for the first machine learning model (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold” Examiner notes that Appel teaches inclusion of subset of features belonging to graphs over a threshold and exclude a subset of features belonging to graphs that do not meet the threshold”) in response to (i) the first machine learning model corresponding to the first task category (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”) and (ii) the first machine learning model being a first machine learning type corresponding to one of a linear regression model, a logistic regression model, a random forest model, or a gradient boosting machine; and (Hamilton, sec. 2.1: “For example, one could feed the learned embeddings to a logistic regression classifier to predict the community that a node belongs to [47], or one could use distances between the embeddings to recommend friendship links in a social network [3, 28] (Section 2.7 discusses further applications).”)
configuring the second machine learning model comprises selecting the second feature subset of the aggregated set of features as input parameters for the second machine learning model (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”) in response to (i) the second machine learning model corresponding to the second task category (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”), and (ii) the second machine learning model being a second machine learning type corresponding to another one of the linear regression model, the logistic regression model, the random forest model, or the gradient boosting machine. (Hamilton, sec. 2.1: “For example, one could feed the learned embeddings to a logistic regression classifier to predict the community that a node belongs to [47], or one could use distances between the embeddings to recommend friendship links in a social network [3, 28] (Section 2.7 discusses further applications).”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claims 4 and 15, Appel as modified teaches:
wherein storing the aggregated node embeddings comprises, with respect to a given node of the graph, storing, in the one or more databases, (Appel, para. 0038: “In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets. In embodiments of the invention, the repository 502 can be implemented as a relational database that is located in memory (e.g., memory 828 shown in FIG. 8) or any other storage location of the system 110.”), a first aggregated node embedding by generating a linear combination of a first node embedding from the first set of node embeddings and a second node embedding from the second set of node embeddings, (Hamilton, pg. 2.3.2: “GCNs and column networks use a weighted sum in line 5 and a (weighted) element-wise mean in line 4” and pg. 14, fig. 8)
wherein the first node embedding and the second node embedding correspond to the same node. (Hamilton, pg. 14, fig. 8: “Figure 8: A, Example of a 4-layer graph, where the same nodes occur in multiple different layers. This multi-layer structure can be exploited to regularize learning at the different layers by requiring that the embeddings for the same node in different layers are similar to each other.” Examiner notes Hamilton at sec. 1.1 teaches both unsupervised and supervised learning for the same node in order to learn different embeddings)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claims 5 and 16, Appel as modified teaches:
Wherein storing the aggregated node embeddings comprises, with respect to a given node of the graph, storing, in the one or more databases, a first aggregated node embedding by averaging a first node embedding from the first set of node embeddings with a second node embedding from the second set of node embeddings, (Hamilton, pg. 2.3.2: “GCNs and column networks use a weighted sum in line 5 and a (weighted) element-wise mean in line 4”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claims 6 and 17, Appel as modified teaches:
The method of claim 2, wherein storing the aggregated node embeddings comprises, with respect to a given node of the graph, storing, in the one or more databases, (Appel, para. 0038: “In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets. In embodiments of the invention, the repository 502 can be implemented as a relational database that is located in memory (e.g., memory 828 shown in FIG. 8) or any other storage location of the system 110.”) a first aggregated node embedding by concatenating (Verma, para. 0114: “To learn the comprehensive node embedding for the root node ‘u’, the processor 206 is configured to fuse/combine/concatenate the direct neighborhood embedding and the skip neighborhood embedding intelligently using an attention module 416.” Examiner notes Verma teaches concatenation of at least a root node with neighboring nodes) a first node embedding derived from using a first embedding model on the given node with a second node embedding derived from using a second embedding model on the given node. (Hamilton, pg. 14, fig. 8: “Figure 8: A, Example of a 4-layer graph, where the same nodes occur in multiple different layers. This multi-layer structure can be exploited to regularize learning at the different layers by requiring that the embeddings for the same node in different layers are similar to each other.” Examiner notes Hamilton at sec. 1.1 teaches both unsupervised and supervised learning for the same node in order to learn embeddings)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel set forth above with respect to claim 1.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claims 7 and 18, Appel as modified teaches:
The method of claim 2, further comprising: after configuring the first machine learning model with the first feature subset, displaying a user interface, comprising an indication of features of the first feature subset of the aggregated set of features, and (Appel, para. 0033: “A user 102 provides inputs to and receives information from the system 110 using, for example, a display (e.g., I/O device 830 of the computing system 830 shown in FIG. 8).”),
removing, from the first machine learning model, a given feature corresponding to a user selection obtained via the user interface, (Appel, para. 0046: “To further refine the selection and evaluation performed by the methodology 300, block 310 optionally presents the user 102 with additional questions on how to map the graph features to the learning task(s) selected. For example, where the task selected is solving a node classification problem, the user 102 can be asked to select how to map node attributes to classes.”)
wherein the first machine learning model no longer uses the given feature for performing tasks of the first task category (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”) after the given feature is removed from the first machine learning model. (Appel, para. 0050: “Block 320 uses the resulting embeddings from block 316 to perform the task(s) selected/identified by the user 102. The results of the task(s) are made available to the user 102.” Examiner notes Appel teaches using resulting embeddings identified by user to perform a task i.e. using the machine learning model where first Appel teaches, above, removing unwanted elements i.e. refining the selection and evaluation of the methodology).
Claims 12 and 21, Appel as modified teaches:
wherein each subset of a plurality of subsets of the aggregated set of features (Appel, para. 0038: “ In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets.”) corresponds to a given machine learning model of a plurality of machine learning models (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”), the operations further comprising:
Training (Appel, para. 0025: “In some embodiments of the invention, the embedding analyzer module can be implemented as a machine learning model that has been trained to perform the above-described mapping and embedding technique selection/recommendation using the graph/embedding data sets in the repository”), based on the plurality of subsets, (Appel, para. 0040: “ In some embodiments of the invention, the above-described operations of the embedding evaluator 116 of the embedding analyzer 114 can be implemented by the set of embedding rules 118 of the embedding analyzer 114, wherein the embedding rules are implemented as a rule-based algorithm configured and arranged to perform the above-described mapping and embedding technique selection/recommendation using the graph/embedding data sets in the repository 502 (shown in FIG. 5).”) the plurality of machine learning models to perform actions (Appel, para. 0035: “In general, the natural language processing algorithms used in the system 110 include speech recognition functionality that allows the system 110 to receive natural language data (text and audio) and apply elements of language processing, information retrieval, and machine learning to derive meaning from the natural language inputs and potentially take action based on the derived meaning”),
wherein each machine learning model of the plurality of machine learning models is trained using the corresponding subset of the plurality of subsets of the aggregated set of features. (Appel, para. 0037: “Classification tasks often depend on the use of labeled datasets to train the neural network to recognize the correlation between labels and data. This is known as supervised learning”)
Claim 13, Appel as modified teaches:
One or more non-transitory, computer-readable media comprising instructions that (Appel, para. 0086: “The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device.”) when executed by one or more processors, causes operations comprising (Appel, para. 0090: “These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions,”)
Storing, for each node of a plurality of nodes of a graph, an aggregated node embedding that (i) is an aggregation of node embeddings respectively derived from different embedding models (Verma, para. 0047: “Furthermore, the present disclosure provides significantly more robust solutions because of handling simultaneous/concurrent processor execution (such as applying one or more neural network models over the same input, simultaneously)” Examiner notes Verma teaches applying different models to same input where Appel teaches an aggregation of embeddings) and (ii) comprises an aggregated set of features with features of a plurality of the derived embeddings; (Appel, para. 0039: “Thus, the embedding technique associated with the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs (i.e., the graph-under-development 104, the extracted features generated by the graph analyzer 112, and the desired task 108) that is over the threshold can be identified or recommended by the embedding analyzer module 114 as an embedding technique for the graph-under-development 104 and the desired task 108.”)
after storing the aggregated set of node embeddings having the aggregated set of features, (Appel, para. 0038: “ In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets.”) configuring a first machine learning model… (Appel, para. 0025: “In some embodiments of the invention, the embedding analyzer module can be implemented as a machine learning model and a set of embedding rules that perform their mapping analyses in series such that the output of the first analysis is fed as an input to the second analysis, thereby improving the overall confidence in the embedding technique selection/recommendation.”)
to incorporate a first feature subset of the aggregated set of features without incorporating one or more features of a second feature subset of the aggregated set of features (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold… Thus, the embedding technique associated with the known graphs, characteristics of the known graphs, and the known overall tasks that have a level of similarity to the inputs that is over the threshold can be identified or recommended by the embedding analyzer module as an embedding technique for the graph-under-development and the desired task.” Examiner notes that for examination purposes only, “incorporating” is interpreted to mean a formal integration of something i.e. something more integral than “including”; examiner notes that Appel teaches inclusion of subset of features belonging to graphs over a threshold and exclude a subset of features belonging to graphs that do not meet the threshold)
after storing the aggregated set of node embeddings having the aggregated set of features, (Appel, para. 0038: “ In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets.”) configuring a second machine learning model corresponding to a second task category (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”) to incorporate the second feature subset (Appel, para. 0030: “The system 110 includes a graph analyzer 112 and an embedding analyzer 114 configured to receive as inputs a graph-under-development 104, along with optionally one or more desired tasks 108. In some embodiments of the invention, the desired task(s) 108 can selected from a set of task options that can be generated by the system 110. The system 110 is communicatively coupled to a configuration of machine learning algorithms 120, although in some embodiments of the invention, the machine learning algorithms 120 can be incorporated within the system 110.”) without incorporating one or more features of the first feature subset by, (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold… Thus, the embedding technique associated with the known graphs, characteristics of the known graphs, and the known overall tasks that have a level of similarity to the inputs that is over the threshold can be identified or recommended by the embedding analyzer module as an embedding technique for the graph-under-development and the desired task.” Examiner notes that for examination purposes only, “incorporating” is interpreted to mean a formal integration of something i.e. something more integral than “including”; examiner notes that Appel teaches inclusion of subset of features belonging to graphs over a threshold and exclude a subset of features belonging to graphs that do not meet the threshold.)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel as modified as set forth above with respect to claim 1.
Claim 19, Appel as modified teaches:
The one or more non-transitory, computer-readable media of claim 13, wherein configuring the first machine learning model comprises selecting the first feature subset of the aggregated set of features as input parameters for the first machine learning model in response to (i) the first machine learning model being the first machine learning type and (Verma, para. 0134: “The regular transaction model 548 is configured to calculate the account intelligence score of everyday spending for each cardholder of the plurality of cardholders 124a-124c.”) (ii) a determination that the first machine learning model would satisfy a performance threshold after training the first machine learning model with the first feature subset of the aggregated set of features. (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel as set forth above with respect to claim 1.
Claim 20, Appel as modified teaches:
The media of claim 13, wherein configuring the first machine learning model comprises selecting the first feature subset of the aggregated set of features as input parameters for the first machine learning model in response to (i) the first machine learning model being the first machine learning type and (Appel, para. 0040: “ In some embodiments of the invention, the above-described operations of the embedding evaluator 116 of the embedding analyzer 114 can be implemented by the set of embedding rules 118 of the embedding analyzer 114, wherein the embedding rules are implemented as a rule-based algorithm configured and arranged to perform the above-described mapping and embedding technique selection/recommendation using the graph/embedding data sets in the repository 502 (shown in FIG. 5).”) (ii) a determination that the first feature subset has features corresponding to a threshold number of eigenvectors derived from a matrix of eigenvectors associated with the aggregated set of node embeddings. (Appel, para. 0043: “The properties of these matrices, especially spectral properties (eigenvalues and eigenvectors) convey information about the structure of the corresponding graph. Thus, a rich view of the properties of graphs can be obtained from the graph's eigenvalues of its adjacency matrix. In general, the set of eigenvalues of a graph G is known as the spectrum of G and denoted by Sp(G).” Examiner notes that Appel teaches a subset of all features of a graph—spectral properties. The spectral properties correspond with the set of eigenvalues of Sp(G) where a threshold number of eigenvectors may be the number of values in the set Sp(G)).
Claim 22, Appel as modified teaches the machine learning system of claim 1:
wherein configuring the first machine learning model comprises selecting the first feature subset as input parameters (Verma, para. 0134: “The regular transaction model 548 is configured to calculate the account intelligence score of everyday spending for each cardholder of the plurality of cardholders 124a-124c.”) for the first machine learning model in response to (i) the first machine learning model corresponding to the first task category (Appel, para. 0037: “Examples of classification tasks include detecting people/faces in images, recognizing facial expressions (e.g., angry, joyful, etc.) in an image, identifying objects in images (e.g., stop signs, pedestrians, lane markers, etc.), recognizing gestures in video, detecting voices, detecting voices in audio, identifying particular speakers, transcribing speech into text, the like.” Examiner notes Hamilton teaches multiple task categories including a first task category), and (ii) a determination that the first machine learning model would satisfy a performance threshold after training the first machine learning model with the first feature subset. (Appel, para. 0024: “The embedding evaluator module maps the above-described inputs against the repository of associated graph/embedding data sets to identify the known graphs, characteristics of the known graphs, and the known overall tasks that have a similarity level with respect to the inputs that is over a threshold” Examiner notes Hamilton teaches a subgraph and Appel teaches features where the features of a subgraph are a feature subset)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel as modified as set forth above with respect to claim 1.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claim 23, Appel as modified teaches the machine learning system of claim 1:
wherein configuring the first machine learning model comprises selecting the first feature subset as input parameters for the first machine learning model in response to (i) the first machine learning model corresponding to the first task category (Appel, para. 0040: “ In some embodiments of the invention, the above-described operations of the embedding evaluator 116 of the embedding analyzer 114 can be implemented by the set of embedding rules 118 of the embedding analyzer 114, wherein the embedding rules are implemented as a rule-based algorithm configured and arranged to perform the above-described mapping and embedding technique selection/recommendation using the graph/embedding data sets in the repository 502 (shown in FIG. 5).” Examiner notes Hamilton teaches multiple task categories including a first task category) and (ii) a determination that the first feature subset has features corresponding to a threshold number of eigenvectors derived from a matrix of eigenvectors associated with the aggregated node embeddings. (Appel, para. 0043: “The properties of these matrices, especially spectral properties (eigenvalues and eigenvectors) convey information about the structure of the corresponding graph. Thus, a rich view of the properties of graphs can be obtained from the graph's eigenvalues of its adjacency matrix. In general, the set of eigenvalues of a graph G is known as the spectrum of G and denoted by Sp(G).” Examiner notes that Appel teaches a subset of all features of a graph—spectral properties. The spectral properties correspond with the set of eigenvalues of Sp(G) where a threshold number of eigenvectors may be the number of values in the set Sp(G) and Hamilton teaches aggregated node embeddings).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Claim 24, Appel as modified teaches the machine learning system of claim 1:
wherein the first feature subset (Appel, para. 0038: “ In accordance with aspects of the invention, the embedding evaluator 116 uses classification and similarity machine learning techniques to generate an embedding/task model 116A (shown in FIG. 5) that can be used to map the inputs received at the embedding evaluator 116 against a graph/embedding repository 502 (shown in FIG. 5) having stored therein associated graph/embedding data sets.”) corresponds to node embeddings derived from the unsupervised embedding model (Appel, para. 0037: “Classification tasks often depend on the use of labeled datasets to train the neural network to recognize the correlation between labels and data. This is known as supervised learning.” Examiner notes Hamilton at sec. 1.1 teaches both unsupervised and supervised learning for the same node in order to learn embeddings) and the second feature subset corresponds to node embeddings derived from the unsupervised embedding model. (Verma, para. 0035: “In addition, the present disclosure uses various aggregation blocks for capturing information (i.e., learning embeddings) from the direct neighbors as well as the skip neighbors of the root node… In other words, the dual loss function captures the topological information of the graph and also enriches the embeddings by maximizing the mutual information. This is achieved by combining a graph structural loss and a mutual information (MI) maximization loss between the node embeddings learned from the neighborhood nodes and self-node features. The BipGNN model is trained in an unsupervised manner so that the embeddings are task agnostic and can be used for multiple downstream tasks.” Examiner notes Hamilton at sec. 1.1 teaches both unsupervised and supervised learning for the same node in order to learn embeddings)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Verma into Appel as set forth above with respect to claim 1.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hamilton into Appel as modified as set forth above with respect to claim 1.
Response to Applicant Remarks/Argument:
Applicant argues that prior art, Appel in view of Verma does not teach the claims as newly amended. However, in light of applicant’s amendments, the claims now stand rejected over Appel in view of Verma and further in view of Hamilton.
Hamilton specifically teaches processing of a same graph node appearing in multiple layers of a graph as well as both supervised and unsupervised learning techniques.
Applicant further argues on page 12 of the remarks that Appel does not teach filtering based on embedding features but rather based on embedding techniques.
However, Applicant’s claims do not specifically require a filtering based on embedding features but rather a performance threshold (claims 19 and 22). In this case, Appel teaches a filtering based on similarity of input on the assumption (i.e. basis that) similar input and data will behave and similarly when input into the same learning model such that the learning model is selected based on an expected performance i.e. meets an expected performance threshold. Therefore, Appel does teach use of “an [unspecific] performance threshold”.
At the bottom of page 12 of applicant’s remarks, Applicant further argues that Verma fails to teach an aggregated set of features that are partitioned into subsets.
However, as set forth in the rejection above, such subsets are taught by Hamilton.
Therefore, for the reasons set forth above, applicant’s claims 1-7 and 12-24 are rejected.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Sally T. Ley whose telephone number is (571)272-3406. The examiner can normally be reached Monday - Thursday, 10:00am - 6:00pm ET.
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, Viker Lamardo can be reached at (571) 270-5871. 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.
/STL/Examiner, Art Unit 2147
/ERIC NILSSON/Primary Examiner, Art Unit 2151
Prosecution Insights