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 .
Specification
The disclosure is objected to because of the following informalities: ¶[0013] refers to ‘feature A’ where the context requires ‘node A’ for consistency with the remainder of the paragraph.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
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.
Claims 18-20 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.
Claim 18 recites "the plurality of nodes" in the limitation "wherein each of the sub-graphs includes a different subset of the plurality of nodes." There is insufficient antecedent basis for this limitation in the claim. Claim 18 never previously recites "a plurality of nodes"; it recites only "nodes of each of a plurality of sub-graphs" and "each of the nodes," neither of which introduces a "plurality of nodes" as an antecedent.
For purposes of examination, "the plurality of nodes" is interpreted under BRI to refer to the complete set of nodes collectively spanning the plurality of sub-graphs, consistent with the specification's teaching at ¶[0075] that the graph network "comprises a plurality of nodes interconnected by a plurality of edges."
Claim 19 depends from claim 18 and therefore incorporates the indefinite "the plurality of nodes" limitation addressed above. Claim 19 is indefinite for at least this reason. See the BRI interpretation provided for claim 18, above.
Claim 19 recites "the prediction" in the limitation "wherein the prediction is generated with respect to an entity corresponding to the point node." There is insufficient antecedent basis for this limitation in the claim. Claim 18, from which claim 19 depends, recites only "one or more outputs representing one or more predictions" — a plural antecedent — and does not introduce a singular "a prediction." It is unclear whether "the prediction" refers to a single one of the "one or more predictions," to all of them collectively, or to some other prediction. For purposes of examination, "the prediction" is interpreted under BRI to mean "each of the one or more predictions" of claim 18, generated with respect to an entity corresponding to the point node of the sub-graph on which that prediction is based.
Claim 20 depends from claim 19, which depends from claim 18, and therefore incorporates the indefinite limitations addressed above with respect to claims 18 and 19. Claim 20 is indefinite for at least these reasons. See the BRI interpretations provided for claims 18 and 19, above.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-10, 12, 15, 16 and 18-20 are rejected under 35 USC 103 as being unpatentable over US Pat. Pub. No. 2024/0062041A1 to Choudhary et al. (hereinafter Choudhary) in view of k-hop Graph Neural Networks to Nikolentzos et al. (hereinafter Nikolentzos).
Per claim 1, Choudhary discloses A method (Choudhary: ¶[0089]…Choudhary discloses a computer-implemented fraud-detection process carried out by the neural network engine of a server system, which constitutes the claimed method under BRI, "In an embodiment, the neural network engine 222 includes suitable logic and/or interfaces for implementing or running the GNN model 232 to perform fraud detection."), comprising:
accessing a graph network of a service provider, wherein the graph network includes a plurality of nodes interconnected by a plurality of edges (Choudhary: ¶[0085]…Choudhary's graph creation engine builds a base graph for a payment-network service provider in which the transacting entities are the nodes and the payment transactions between them are the edges, which constitutes a graph network of a service provider having a plurality of nodes interconnected by a plurality of edges under BRI, "The base graph represents a computer-based graph representation of the plurality of entities 104a-104c as nodes...In addition, relationships between the nodes are represented as edges. The edges represent payment transactions performed between the plurality of payment instruments");
generating a plurality of sub-graphs, wherein each of the sub-graphs corresponds to a different portion of the graph network, and wherein each of the sub-graphs includes a different subset of the plurality of nodes (Choudhary: ¶[0093]…Choudhary divides the base graph into three sub-graphs that each contain the node ‘n’ together with a different subset of its neighboring nodes, which constitutes generating a plurality of sub-graphs each corresponding to a different portion of the graph network and each containing a different subset of the nodes under BRI, "In this example scenario, the sub-graph creation engine 224 is configured to segment or divide the base graph into three sub-graphs…A first sub-graph of the three sub-graphs (e.g., sub-graph 1) may include the node ‘n’ and all its neighboring nodes that are labeled as fraudulent. A second sub-graph of the three sub-graphs (e.g., sub-graph 2) may include the node ‘n’ and all its neighboring nodes that are labeled as non-fraudulent. A third sub-graph of the three sub-graphs (e.g., sub-graph 3) may include the node ‘n’ and all its neighboring nodes that are unlabeled"); …
generating one or more predictions utilizing the trained GNN model (Choudhary: ¶[0089]…Choudhary's trained GNN model outputs a fraudulent or non-fraudulent label for each previously unlabeled node, which constitutes generating one or more predictions utilizing the trained GNN model under BRI, "In particular, the neural network engine 222 is configured to identify labels for the unlabeled nodes of the base graph. More specifically, the neural network engine 222 is configured to determine whether a particular node of the base graph is fraudulent or non-fraudulent").
Choudhary does not expressly disclose, but Nikolentzos does teach:
defining a directional flow for information exchanges between the nodes of each of the sub-graphs (Nikolentzos: p. 7, Section 4.1…Nikolentzos organizes each rooted sub-graph into rings of nodes indexed by their hop distance from the root and then propagates node representations inward, ring by ring, from the most distant nodes toward the root, which constitutes defining a directional flow for the information exchanges between the nodes of each sub-graph under BRI, "The proposed approach starts from the most distant nodes and follows a sequential procedure updating the feature vectors of nodes that are gradually closer to the root”; “R1 (v) = N1 (v) is the set of direct neighbors of v, while Rd (v) denotes the ring of nodes at distance d, which we refer to as the nodes at level d");
training a graph neural network (GNN) model based on the defined directional flow (Nikolentzos: p. 6, Section 4.1…Nikolentzos executes the neighborhood-aggregation layers of its k-hop GNN as an ordered series of UPDATE modules that follow that inward sequence, so the resulting model is trained on the defined directional flow, which constitutes the claimed training step under BRI, "The proposed approach uses a series of UPDATE modules to update the representations of the nodes that belong to the k-hop neighborhood of v, following a sequential procedure from the most distant ones to the direct neighbors of v").
Choudhary and Nikolentzos are analogous art because they are from the same field of endeavor, specifically graph neural networks that learn node representations by aggregating information across the neighborhood of a node in a graph. They are further reasonably pertinent to the same problem of improving the accuracy of node-level predictions produced by a graph neural network trained on neighborhood sub-graphs.
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to govern the message passing in the neighborhood sub-graphs of Choudhary's fraud-detection graph neural network by the hop-distance-ordered, inward-directed update sequence taught by Nikolentzos. This is the use of a known technique to improve a similar device in the same way (KSR rationale (C)), and it yields no more than the predictable result of a GNN whose node representations are aggregated in a defined order rather than indiscriminately.
The suggestion/motivation for doing so would have been provided by Nikolentzos itself, which teaches that aggregating over the whole k-hop neighborhood in this ordered fashion yields a more expressive model that captures structural information a standard neighbor-aggregating GNN cannot, "we propose a more expressive architecture, k-hop GNNs, which updates a node’s representation by aggregating information not only from its direct neighbors, but from its k-hop neighborhood." (Nikolentzos: p. 1, Abstract), and reports that "the proposed model achieves performance better or comparable to standard GNNs and to state-of-the-art algorithms." (Nikolentzos: p. 1, Abstract). Choudhary expressly seeks that same improvement, stating that "For the node classification task, the probability of correctly classifying a node can be enhanced by increasing the positive ratio of each node" (Choudhary: ¶[0121]), so a PHOSITA would have been motivated to adopt Nikolentzos's ordered aggregation to raise the accuracy of Choudhary's fraud/non-fraud node classification.
Per claim 2, Choudhary combined with Nikolentzos discloses claim 1. Choudhary further teaches: the plurality of sub-graphs is generated such that each of the sub-graphs includes a point node, respectively (Choudhary: ¶[0092]…Choudhary builds each of its sub-graphs around a single node ‘n’ that is carried into every sub-graph, which constitutes each sub-graph including a point node, respectively, under BRI, "In one example, the base graph is a computer-based graph representation of a node ‘n’ and all its neighboring nodes i.e., nodes that are directly connected with the node ‘n’").
Nikolentzos further teaches the defining the directional flow is based on distances between the point node and a rest of the nodes in each of the sub-graphs (Nikolentzos: p. 7, Section 4.1…Nikolentzos computes the hop-count distance from the root to every other node in the rooted sub-graph and uses those distances to order the inward update sequence, which constitutes defining the directional flow based on distances between the point node and the rest of the nodes under BRI, "Let Rd (v) denote the set of nodes at distance (hop count) exactly d > 0 from v…The proposed approach starts from the most distant nodes and follows a sequential procedure updating the feature vectors of nodes that are gradually closer to the root").
The rationale to combine Nikolentzos with Choudhary is the same as the parent claim.
Per claim 3, Choudhary combined with Nikolentzos discloses claim 2. Choudhary further teaches performing one or more preprocesses to the graph network before the generating of the plurality of sub-graphs, wherein performing the preprocesses comprises: … embedding one or more data features in each of the nodes in each of the sub-graphs (Choudhary: ¶[0085], ¶[0087]…Choudhary extracts graph features from the transaction data and embeds them, together with per-node metadata, into the nodes of the base graph before the sub-graph creation engine divides that graph, which constitutes performing a preprocess of embedding one or more data features in each of the nodes before the sub-graphs are generated under BRI, "In an embodiment, the graph creation engine 220 includes suitable logic and/or interfaces for defining or generating the base graph based, at least in part, on the plurality of graph features identified from the electronic transaction data…Additionally, the base graph may include metadata associated with the plurality of nodes, and/or information identifying relationships (such as, for example, electronic transactions, fraud connections, etc.) among the plurality of nodes").
Per claim 4, Choudhary combined with Nikolentzos discloses claim 2. Nikolentzos further teaches the directional flow is defined such that the information exchanges between a subset of the nodes are uni-directional toward the point node (Nikolentzos: p. 7, Section 4.1…in Nikolentzos's across-ring update a node at level d aggregates only from its directly connected neighbors at level d+1, so for that subset of node pairs the information travels one way only, inward toward the root, which constitutes information exchanges between a subset of the nodes being uni-directional toward the point node under BRI, "Let also B = N1 (u) ∩ Rd+1 (v) denote the neighbors of u that belong to level d + 1 of Gkv…The proposed approach starts from the most distant nodes and follows a sequential procedure updating the feature vectors of nodes that are gradually closer to the root"). The rationale to combine Nikolentzos with Choudhary is the same as the parent claim.
Per claim 5, Choudhary combined with Nikolentzos discloses claim 2. Nikolentzos further teaches the directional flow is defined at least in part based on a comparison of a first distance between the point node and a first node of the plurality of nodes and a second distance between the point node and a second node of the plurality of nodes (Nikolentzos: p. 7, Section 4.1…Nikolentzos selects the sending nodes by comparing their level index d+1 against the receiving node's level index d, both measured as distances from the root, which constitutes defining the directional flow at least in part on a comparison of a first and a second distance to the point node under BRI, "Let also B = N1 (u) ∩ Rd+1 (v) denote the neighbors of u that belong to level d + 1 of Gkv"; "R1 (v) = N1 (v) is the set of direct neighbors of v, while Rd (v) denotes the ring of nodes at distance d, which we refer to as the nodes at level d"). The rationale to combine Nikolentzos with Choudhary is the same as the parent claim.
Per claim 6, Choudhary combined with Nikolentzos discloses claim 2. Choudhary further teaches the one or more predictions are generated with respect to the point node (Choudhary: ¶[0120]…Choudhary generates its fraud or non-fraud label for the particular node under consideration around which the sub-graphs were built, which constitutes generating the one or more predictions with respect to the point node under BRI, "The processor 206 is then configured to split the base graph 502 into label-aware neighborhoods Nk(v), where v represents an example node under consideration i.e., v ∈ {unknown, fraud, non-fraud} and k is its label").
Per claim 7, Choudhary combined with Nikolentzos discloses claim 1. Nikolentzos further teaches the directional flow is defined for the information exchanges between different pairs of directly-connected nodes in each of the sub-graphs (Nikolentzos: p. 7, Section 4.1…Nikolentzos restricts each update to the set N1(u) of nodes directly connected to u, so the direction is fixed pair-by-pair for every directly connected pair in the sub-graph, which constitutes defining the directional flow for exchanges between different pairs of directly-connected nodes under BRI, "Let also B = N1 (u) ∩ Rd+1 (v) denote the neighbors of u that belong to level d + 1 of Gkv ."; "After all its direct neighbors u ∈ N1 (v) have been processed, the hidden state of the root node v is computed as follows:"). The rationale to combine Nikolentzos with Choudhary is the same as the parent claim.
Per claim 8, Choudhary combined with Nikolentzos discloses claim 1. Choudhary further teaches each node of the plurality of nodes is associated with a respective user account with the service provider; and each edge of the plurality of edges is associated with an interaction between the respective user accounts associated with the nodes that are interconnected by the edge (Choudhary: ¶[0107]…Choudhary's nodes are the payment accounts held with the payment service provider and each edge is the electronic transaction conducted between the two accounts it joins, which constitutes each node being associated with a respective user account and each edge with an interaction between those accounts under BRI, "In addition, the nodes may represent payment instruments including, for example, payment accounts, payment cards, payment wallets, and the like"; ¶[0120]…"The edges between the nodes represent the electronic transactions performed between the two nodes (e.g., two parties or two payment accounts, etc.)").
Per claim 9, Choudhary combined with Nikolentzos discloses claim 1. Choudhary further teaches the one or more predictions comprise a prediction with respect to a predefined activity, a predefined metric, or a predefined decision (Choudhary: ¶[0089]…Choudhary's output is a determination of whether the node is fraudulent, and fraud is a predefined activity, which constitutes a prediction with respect to a predefined activity under BRI, "In particular, the neural network engine 222 is configured to identify labels for the unlabeled nodes of the base graph. More specifically, the neural network engine 222 is configured to determine whether a particular node of the base graph is fraudulent or non-fraudulent.").
Per claim 10, Choudhary combined with Nikolentzos discloses claim 9. Choudhary further teaches the predefined activity comprises an occurrence of fraud… (Choudhary: ¶[0120]…Choudhary assigns each node under consideration to the fraud class, which constitutes the predefined activity comprising an occurrence of fraud under BRI, "The processor 206 is then configured to split the base graph 502 into label-aware neighborhoods Nk(v), where v represents an example node under consideration i.e., v ∈ {unknown, fraud, non-fraud} and k is its label").
Per claim 12, Choudhary discloses A system (Choudhary: ¶[0076]…Choudhary's server system is built on a hardware processor, which constitutes the claimed system under BRI, "Examples of the processor 206 include, but are not limited to, an application-specific integrated circuit (ASIC) processor, a reduced instruction set computing (RISC) processor, a complex instruction set computing (CISC) processor, a field-programmable gate array (FPGA), and the like... In another embodiment, the memory 208 may be realized in the form of a database server or cloud storage working in conjunction with the server system 200"), comprising:
one or more processors (Choudhary: ¶[0076]…Choudhary's processor 206 is a hardware processor such as an ASIC, RISC, CISC or FPGA device, which constitutes one or more processors under BRI, "Examples of the processor 206 include, but are not limited to, an application-specific integrated circuit (ASIC) processor, a reduced instruction set computing (RISC) processor, a complex instruction set computing (CISC) processor, a field-programmable gate array (FPGA), and the like"); and
a non-transitory computer-readable medium having stored thereon instructions that are executable by the one or more processors that cause the system to perform operations comprising (Choudhary: ¶[0076]…Choudhary's memory 208 is a tangible RAM, ROM, removable drive or hard disk that stores the computer-readable instructions the processor executes, which constitutes a non-transitory computer-readable medium storing processor-executable instructions under BRI, "The memory 208 includes suitable logic, circuitry, and/or interfaces to store a set of computer-readable instructions for performing operations. Examples of the memory 208 include a random-access memory (RAM), a read-only memory (ROM), a removable storage drive, a hard disk drive (HDD), and the like"):
accessing a graph that includes a plurality of nodes that are interconnected, wherein each of the nodes represents a different entity (Choudhary: ¶[0085]…Choudhary's base graph interconnects nodes by edges and each node stands for a distinct transacting entity, which constitutes accessing a graph of interconnected nodes each representing a different entity under BRI, "The base graph represents a computer-based graph representation of the plurality of entities 104a-104c as nodes…In addition, relationships between the nodes are represented as edges. The edges represent payment transactions performed between the plurality of payment instruments");
dividing the graph into a plurality of sub-graphs, wherein each of the sub-graphs includes a different subset of the plurality of nodes, and wherein each of the sub-graphs includes a point node, respectively (Choudhary: ¶[0092]…Choudhary segments the base graph into three sub-graphs, each holding a different subset of the neighbors and each built around the same node ‘n’, which constitutes dividing the graph into sub-graphs that each include a different subset of the nodes and each include a point node under BRI, "In one example, the base graph is a computer-based graph representation of a node ‘n’ and all its neighboring nodes i.e., nodes that are directly connected with the node ‘n’."; ¶[0093]…"In this example scenario, the sub-graph creation engine 224 is configured to segment or divide the base graph into three sub-graphs…A first sub-graph of the three sub-graphs (e.g., sub-graph 1) may include the node ‘n’ and all its neighboring nodes that are labeled as fraudulent. A second sub-graph of the three sub-graphs (e.g., sub-graph 2) may include the node ‘n’ and all its neighboring nodes that are labeled as non-fraudulent. A third sub-graph of the three sub-graphs (e.g., sub-graph 3) may include the node ‘n’ and all its neighboring nodes that are unlabeled");…
generating one or more predictions via the trained GNN model (Choudhary: ¶[0089]…Choudhary's trained GNN model outputs a fraudulent or non-fraudulent label for each previously unlabeled node, which constitutes generating one or more predictions utilizing the trained GNN model under BRI, "In particular, the neural network engine 222 is configured to identify labels for the unlabeled nodes of the base graph. More specifically, the neural network engine 222 is configured to determine whether a particular node of the base graph is fraudulent or non-fraudulent").
Choudhary does not expressly disclose, but Nikolentzos does teach:
determining, for each of the sub-graphs, distances between the point node and a rest of the nodes in the sub-graph (Nikolentzos: p. 7, Section 4.1…Nikolentzos determines, for the rooted sub-graph around each root node, the exact hop count separating the root from every other node in that sub-graph, which constitutes determining, for each of the sub-graphs, distances between the point node and the rest of the nodes under BRI, "Let Rd (v) denote the set of nodes at distance (hop count) exactly d > 0 from v…The proposed approach starts from the most distant nodes and follows a sequential procedure updating the feature vectors of nodes that are gradually closer to the root"; “R1 (v) = N1 (v) is the set of direct neighbors of v, while Rd (v) denotes the ring of nodes at distance d, which we refer to as the nodes at level d");
training a graph neural network (GNN) model based on a directional flow of information among the nodes in each of the sub-graphs, wherein the directional flow of information is defined at least in part based on the determined distances between the point node and the rest of the nodes in the sub-graph (Nikolentzos: p. 6-7, Section 4.1…Nikolentzos trains its k-hop GNN with a series of UPDATE modules that move information inward from the most distant ring to the root, the ring membership being fixed by the measured hop distances, which constitutes training a GNN model on a directional flow defined at least in part by the determined distances under BRI, "The proposed approach uses a series of UPDATE modules to update the representations of the nodes that belong to the k-hop neighborhood of v, following a sequential procedure from the most distant ones to the direct neighbors of v"; "The proposed approach starts from the most distant nodes and follows a sequential procedure updating the feature vectors of nodes that are gradually closer to the root”).
Choudhary and Nikolentzos are analogous art because they are from the same field of endeavor, specifically graph neural networks that learn node representations by aggregating information across the neighborhood of a node in a graph. They are further reasonably pertinent to the same problem of improving the accuracy of node-level predictions produced by a graph neural network trained on neighborhood sub-graphs.
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to govern the message passing in the neighborhood sub-graphs of Choudhary's fraud-detection graph neural network by the hop-distance-ordered, inward-directed update sequence taught by Nikolentzos. This is the use of a known technique to improve a similar device in the same way (KSR rationale (C)), and it yields no more than the predictable result of a GNN whose node representations are aggregated in a defined order rather than indiscriminately.
The suggestion/motivation for doing so would have been provided by Nikolentzos itself, which teaches that aggregating over the whole k-hop neighborhood in this ordered fashion yields a more expressive model that captures structural information a standard neighbor-aggregating GNN cannot, "we propose a more expressive architecture, k-hop GNNs, which updates a node’s representation by aggregating information not only from its direct neighbors, but from its k-hop neighborhood." (Nikolentzos: p. 1, Abstract), and reports that "the proposed model achieves performance better or comparable to standard GNNs and to state-of-the-art algorithms." (Nikolentzos: p. 1, Abstract). Choudhary expressly seeks that same improvement, stating that "For the node classification task, the probability of correctly classifying a node can be enhanced by increasing the positive ratio of each node" (Choudhary: ¶[0121]), so a PHOSITA would have been motivated to adopt Nikolentzos's ordered aggregation to raise the accuracy of Choudhary's fraud/non-fraud node classification.
Per claim 15, Choudhary combined with Nikolentzos discloses claim 12. Choudhary further teaches the one or more predictions are generated with respect to the point node (Choudhary: ¶[0120]…Choudhary generates its fraud or non-fraud label for the particular node under consideration around which the sub-graphs were built, which constitutes generating the one or more predictions with respect to the point node under BRI, "The processor 206 is then configured to split the base graph 502 into label-aware neighborhoods Nk(v), where v represents an example node under consideration i.e., v ∈ {unknown, fraud, non-fraud} and k is its label").
Per claim 16, Choudhary combined with Nikolentzos discloses claim 15. Choudhary further teaches the graph comprises a transaction graph of a service provider; the plurality of nodes represent a plurality of users of the service provider; and the point node represents a user that is associated with a fraudulent activity (Choudhary: ¶[0085]…Choudhary's base graph is a transaction graph maintained by a payment service provider whose nodes are the payment accounts of its users, and the node under consideration is classified as belonging to the fraud class, which constitutes a transaction graph of a service provider whose nodes represent users and whose point node represents a user associated with a fraudulent activity under BRI, "the graph creation engine 220 includes suitable logic and/or interfaces for defining or generating the base graph based, at least in part, on the plurality of graph features identified from the electronic transaction data. In one non-limiting example, the base graph is a homogeneous graph. The base graph represents a computer-based graph representation of the plurality of entities 104a-104c as nodes"; ¶[0107]…"In addition, the nodes may represent payment instruments including, for example, payment accounts, payment cards, payment wallets, and the like"; ¶[0120]… "The processor 206 is then configured to split the base graph 502 into label-aware neighborhoods Nk(v), where v represents an example node under consideration i.e., v ∈ {unknown, fraud, non-fraud} and k is its label").
Per claim 18, Choudhary discloses A non-transitory machine-readable medium having stored thereon machine-readable instructions executable to cause a machine to perform operations comprising (Choudhary: paragraph [0076]…Choudhary's memory 208 is a tangible RAM, ROM, removable drive or hard disk holding the instruction set the server executes, which constitutes a non-transitory machine-readable medium storing machine-readable instructions under BRI, "The memory 208 includes suitable logic, circuitry, and/or interfaces to store a set of computer-readable instructions for performing operations. Examples of the memory 208 include a random-access memory (RAM), a read-only memory (ROM), a removable storage drive, a hard disk drive (HDD), and the like"):
accessing a graph neural network (GNN)…wherein each of the nodes are interconnected by a plurality of edges in a graph network of a service provider, wherein each of the sub-graphs corresponds to a different portion of the graph network, and wherein each of the sub-graphs includes a different subset of the plurality of nodes (Choudhary: ¶[0085]…Choudhary's nodes are joined by transaction edges within the payment provider's base graph, and its three sub-graphs each cover a different portion of that graph with a different subset of the nodes, which constitutes the recited graph-network and sub-graph structure under BRI, "The base graph represents a computer-based graph representation of the plurality of entities 104a-104c as nodes...In addition, relationships between the nodes are represented as edges. The edges represent payment transactions performed between the plurality of payment instruments"; ¶[0093]… "In this example scenario, the sub-graph creation engine 224 is configured to segment or divide the base graph into three sub-graphs…A first sub-graph of the three sub-graphs (e.g., sub-graph 1) may include the node ‘n’ and all its neighboring nodes that are labeled as fraudulent. A second sub-graph of the three sub-graphs (e.g., sub-graph 2) may include the node ‘n’ and all its neighboring nodes that are labeled as non-fraudulent. A third sub-graph of the three sub-graphs (e.g., sub-graph 3) may include the node ‘n’ and all its neighboring nodes that are unlabeled"); and
generating, using the trained GNN model, one or more outputs representing one or more predictions associated with a transaction or an offer (Choudhary: ¶[0089]…Choudhary's trained GNN emits a fraud or non-fraud label for a node whose edges are the electronic transactions it participated in, which constitutes generating outputs representing predictions associated with a transaction under BRI, "In particular, the neural network engine 222 is configured to identify labels for the unlabeled nodes of the base graph. More specifically, the neural network engine 222 is configured to determine whether a particular node of the base graph is fraudulent or non-fraudulent"; ¶[0120]… "The edges between the nodes represent the electronic transactions performed between the two nodes (e.g., two parties or two payment accounts, etc.)").
Choudhary does not expressly disclose, but Nikolentzos does teach:
accessing a graph neural network (GNN) model trained based on a directional flow defined for information exchanges between nodes of each of a plurality of sub-graphs (Nikolentzos: p. 6-7, Section 4.1…Nikolentzos's k-hop GNN is trained by propagating information inward through the rooted sub-graph of each node, from the most distant ring to the root, and the resulting model is then applied to node-level tasks, which constitutes accessing a GNN model trained on a directional flow defined for information exchanges between the nodes of each of a plurality of sub-graphs under BRI, "The proposed approach uses a series of UPDATE modules to update the representations of the nodes that belong to the k-hop neighborhood of v, following a sequential procedure from the most distant ones to the direct neighbors of v"; "After T iterations (i.e., T neighborhood aggregation layers), the emerging node feature vectors hv(T) can be used in any node-related task").
Choudhary and Nikolentzos are analogous art because they are from the same field of endeavor, specifically graph neural networks that learn node representations by aggregating information across the neighborhood of a node in a graph. They are further reasonably pertinent to the same problem of improving the accuracy of node-level predictions produced by a graph neural network trained on neighborhood sub-graphs.
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to govern the message passing in the neighborhood sub-graphs of Choudhary's fraud-detection graph neural network by the hop-distance-ordered, inward-directed update sequence taught by Nikolentzos. This is the use of a known technique to improve a similar device in the same way (KSR rationale (C)), and it yields no more than the predictable result of a GNN whose node representations are aggregated in a defined order rather than indiscriminately.
The suggestion/motivation for doing so would have been provided by Nikolentzos itself, which teaches that aggregating over the whole k-hop neighborhood in this ordered fashion yields a more expressive model that captures structural information a standard neighbor-aggregating GNN cannot, "we propose a more expressive architecture, k-hop GNNs, which updates a node’s representation by aggregating information not only from its direct neighbors, but from its k-hop neighborhood." (Nikolentzos: p. 1, Abstract), and reports that "the proposed model achieves performance better or comparable to standard GNNs and to state-of-the-art algorithms." (Nikolentzos: p. 1, Abstract). Choudhary expressly seeks that same improvement, stating that "For the node classification task, the probability of correctly classifying a node can be enhanced by increasing the positive ratio of each node" (Choudhary: ¶[0121]), so a PHOSITA would have been motivated to adopt Nikolentzos's ordered aggregation to raise the accuracy of Choudhary's fraud/non-fraud node classification.
Per claim 19, Choudhary combined with Nikolentzos discloses claim 18. Choudhary further teaches each of the sub-graphs includes a point node, respectively, wherein the prediction is generated with respect to an entity corresponding to the point node (Choudhary:¶[0092]…Choudhary builds every sub-graph around the node ‘n’ and produces its fraud/non-fraud prediction for the entity that node represents, which constitutes each sub-graph including a point node and the prediction being generated with respect to the entity corresponding to it under BRI, "In one example, the base graph is a computer-based graph representation of a node ‘n’ and all its neighboring nodes i.e., nodes that are directly connected with the node ‘n’"; ¶[0120]…"The processor 206 is then configured to split the base graph 502 into label-aware neighborhoods Nk(v), where v represents an example node under consideration i.e., v ∈ {unknown, fraud, non-fraud} and k is its label").
Per claim 20, Choudhary combined with Nikolentzos discloses claim 19. Nikolentzos further teaches defining an information flow direction within each sub-graph at least in part based on distances between the point node and a rest of the nodes in the sub-graph (Nikolentzos: p. 7, Section 4.1…Nikolentzos fixes the direction of every update inside each rooted sub-graph by the hop distances separating the root from the other nodes, which constitutes defining an information flow direction within each sub-graph at least in part based on those distances under BRI, "Let Rd (v) denote the set of nodes at distance (hop count) exactly d > 0 from v…The proposed approach starts from the most distant nodes and follows a sequential procedure updating the feature vectors of nodes that are gradually closer to the root"; "Let also B = N1 (u) ∩ Rd+1 (v) denote the neighbors of u that belong to level d + 1 of Gkv"). The rationale to combine Nikolentzos with Choudhary is the same as the parent claim.
Claims 11 and 17 are rejected under 35 USC 103 as being unpatentable over Choudhary in view of Nikolentzos, as applied in the rejection of claims 1 and 12 above, and further in view of One Trillion Edges: Graph Processing at Facebook-Scale to Ching et al. (hereinafter Ching).
Per claim 11, Choudhary combined with Nikolentzos discloses claim 1.
Choudhary combined with Nikolentzos does not expressly disclose, but with Ching does teach:
the accessing the graph network comprises retrieving the graph network from a Hadoop Distributed File System (HDFS) (Ching: p. 1806, Section 3.2.1…Ching's Giraph graph-processing platform reads the graph it operates on - its vertices and its edges - out of external storage, and expressly identifies HDFS files as the data source it reads that graph from, which constitutes accessing the graph network by retrieving it from a Hadoop Distributed File System under BRI, "Since Giraph is a computing platform, it needs to interface with external storage to read the input and write back the output of a batch computation. Similarly to MapReduce, we can define custom input and output formats for various data sources (e.g., HDFS files, HBase tables, Hive tables)…Datasets fed to a Giraph job consist of vertices and edges, typically with some attached metadata").
Per claim 17, Choudhary combined with Nikolentzos discloses claim 12. Choudhary further teaches the plurality of nodes are interconnected by a plurality of edges that represent interactions among the plurality of nodes (Choudhary: ¶[0120]…Choudhary's nodes are joined by edges that each stand for an electronic transaction conducted between the two parties or payment accounts they connect, which constitutes edges representing interactions among the nodes under BRI, "The edges between the nodes represent the electronic transactions performed between the two nodes (e.g., two parties or two payment accounts, etc.)").
Choudhary combined with Nikolentzos does not expressly disclose, but with Ching does teach:
information associated with the plurality of nodes and the plurality of edges are stored in a Hadoop Distributed File System (HDFS) (Ching: p. 1806, Section 3.1 …Ching holds the vertex and edge datasets of its production graph in HDFS and loads them from there for processing, which constitutes information associated with the nodes and the edges being stored in a Hadoop Distributed File System under BRI, "Giraph directly interfaces with our internal version of HDFS (since Giraph is written in Java) as well as Hive"; p. 1806, Section 3.2.1…"Datasets fed to a Giraph job consist of vertices and edges, typically with some attached metadata").
As it pertains to claims 11 and 17, Choudhary, Nikolentzos and Ching are analogous art because all three are from the same field of endeavor, specifically the large-scale processing of graph-structured data in which information is propagated between connected nodes. Ching is further reasonably pertinent to the particular problem with which the inventor was involved, namely storing and retrieving a graph too large to reside on a single machine so that it can be processed for machine-learning purposes.
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to hold the nodes and edges of Choudhary's payment-transaction base graph in the Hadoop Distributed File System taught by Ching, and to retrieve that graph from HDFS when the graph is accessed for sub-graph generation and graph neural network training. This is the use of a known technique to improve a similar device in the same way (KSR rationale (C)), and it yields no more than the predictable result of a very large transaction graph being held in, and read from, a distributed file system built for graphs of that scale.
The suggestion/motivation for doing so would have been provided by Ching itself, which teaches that a graph-processing platform needs to interface external storage, "Since Giraph is a computing platform, it needs to interface with external storage to read the input and write back the output of a batch computation. Similarly to MapReduce, we can define custom input and output formats for various data sources (e.g., HDFS files, HBase tables, Hive tables)" (Ching: p. 1806, Section 3.2.1), and that interfacing directly with HDFS lets the platform minimize operational overhead, "Since Giraph is scheduled as a MapReduce job, we can leverage our existing MapReduce (Corona) infrastructure stack with little operational overhead" (Ching: p. 1806, Section 3.1). Choudhary's base graph is assembled from the historical payment transaction data of an entire payment network and is therefore of exactly the scale Ching addresses, so a PHOSITA would have been motivated to adopt Ching's HDFS-backed vertex and edge storage to obtain that scalability and operational economy for Choudhary's graph.
Claims 13 and 14 are rejected under 35 USC 103 as being unpatentable over Choudhary in view of Nikolentzos, and further in view of Analyzing Learned Molecular Representations for Property Prediction to Yang et al. (hereinafter Yang).
Per claim 13, Choudhary combined with Nikolentzos discloses claim 12.
Choudhary combined with Nikolentzos does not expressly disclose, but with Yang does teach:
the directional flow of information is defined such that the information flows from a first node of a sub-graph to a second node of the sub-graph only when: the first node and the second node are directly connected; and a distance from the second node to the point node is less than a distance between the first node and the point node (Yang: p. 7, Directed MPNN…Yang carries its messages on directed edges and expressly bars a message from being propagated back to the node it came from, so that a message travels from one node to a directly bonded neighbor in one direction only and never in the reverse direction; applying that exclusivity to Nikolentzos's ring-ordered sub-graph confines every exchange to a directly connected pair in which the receiving node is strictly closer to the root, which constitutes the "only when" condition under BRI, "Using Figure 1 as an illustration, in D-MPNN, the message 1 → 2 will only be propagated to nodes 3 and 4 in the next iteration, whereas in the original MPNN it will be sent to node 1 as well, creating an unnecessary loop in the message passing trajectory…Observe that message mvmt+1 does not depend on its reverse message mwvt from the previous iteration").
Per claim 14, Choudhary combined with Nikolentzos discloses claim 12.
Choudhary combined with Nikolentzos does not expressly disclose, but with Yang does teach:
the training is performed for a plurality of cycles, and wherein in each cycle of the plurality of cycles, information among the nodes flows uni-directionally toward the point node (Yang: p. 7…Directed MPNN…Yang repeats its message passing over successive iterations and, because each message excludes the reverse message from the preceding iteration, the flow across every iteration remains one-directional; applied to Nikolentzos's rooted sub-graph, whose aggregation layers are repeated T times, this yields a training run of multiple cycles in each of which information moves uni-directionally toward the root, which constitutes the "does not depend on its reverse message" limitation under BRI, "Rather than using messages associated with vertices (atoms), D-MPNN uses messages associated with directed edges (bonds)").
As it pertains to claims 13 and 14, Choudhary, Nikolentzos and Yang are analogous art because all three are from the same field of endeavor, specifically neural networks that operate on graph-structured data by passing messages between connected nodes. Yang is further reasonably pertinent to the same problem of suppressing redundant message exchanges between neighboring nodes during graph neural network training.
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to restrict the ring-ordered message passing of Choudhary combined with Nikolentzos to a single direction, as taught by Yang, so that information passes only to a directly connected node that is strictly nearer the point node. This is the use of a known technique to improve a similar device in the same way (KSR rationale (C)).
The suggestion/motivation for doing so would have been provided by Yang itself, which teaches that permitting a message to return along the path it arrived on "in D-MPNN, the message 1 → 2 will only be propagated to nodes 3 and 4 in the next iteration, whereas in the original MPNN it will be sent to node 1 as well, creating an unnecessary loop in the message passing trajectory" (Yang: p. 7, Directed MPNN), and that such returning excursions "are likely to introduce noise into the graph representation" (Yang: p. 7, Directed MPNN). A PHOSITA seeking to reduce the redundant exchanges that Nikolentzos's within-ring updates would otherwise permit would have looked to Yang's directed-edge formulation and confined the flow to strictly distance-decreasing, directly connected node pairs.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALAN CHEN whose telephone number is (571) 272-4143. The examiner can normally be reached M-F 10-7.
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, Kamran Afshar can be reached at (571) 272-7796. 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.
/ALAN CHEN/Primary Examiner, Art Unit 2125
Prosecution Insights