DETAILED ACTION
This action is responsive to Applicant’s reply filed March 25th 2026. This action is made final.
Status of the Claims
Claims 1, 8 and 15 are amended.
Claim status is currently pending and under examination for Claims 1, 4, 6-8, 11, 13-15, 17, 19-22, 26-27, 29-32 of which independent claims are 1, 8 and 15.
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 .
Response to Amendment
Applicant’s arguments regarding the art rejections are moot in view of the new grounds of rejection necessitated by applicant’s amendment.
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 (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 following are the references relied upon in the rejections below:
Zhang, Hao, Zhan Li, and Zhixing Ren. "Data-driven construction of data center graph of things for anomaly detection." arXiv preprint arXiv:2004.12540 (2020).
Taslakian (US 20230025826 A1)
Claims 1, 7-8, 14-15, 20, 29 and 31 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Taslakian.
With respect to claim 1, Zhang teaches:
A method comprising (Zhang discloses “This work proposes a data-driven pipeline (ts2graph) to build a DC graph of things (sensor graph) from the time series measurements of sensors” (P. 1, Abstract).):
ingesting data from an external sensor tracking an industrial process (Zhang discloses “We use the sensory data for a data center building at Alibaba Cloud. The sensors include IT load, real-time power usage effectiveness (PUE), outside environment sensors (6 sensors), water cooling system sensors (25 sensors), hot aisle temperature sensors (58 sensors), and computer room air conditioning (CRAC) unit fan speeds (72 sensors), for a total of 163 sensors. We follow the ts2graph pipeline, and construct the sensor graph. The time series measurement frequency is five minutes.” (P. 8, Sec. 3.1, ¶3).);
incorporating the ingested data into a knowledge graph comprising nodes interconnected by edges (Zhang discloses “This work proposes a data-driven pipeline (ts2graph) to build a DC graph of things (sensor graph) from the time series measurements of sensors. The sensor graph is an undirected weighted property graph, where sensors are the nodes, sensor features are the node properties, and sensor connections are the edges. The sensor node property is defined by features that characterize the sensor events (behaviors), instead of the original time series. The sensor connection (edge weight) is defined by the probability of concurrent events between two sensors” (P. 1, Abstract).),
the nodes representing entities related to the industrial process and the edges represent respective relationships between the entities (Zhang discloses “where the sensor is the node, sensor feature is the node property, and sensor connection is the edge. The connection between two sensors is defined by their concurrent events (events that happen at about the same time). If two sensors share many common events, then they have strong connections” (P. 2, Sec. 1, Last Paragraph).);
training a graph neural network with the knowledge graph to form a reference graph model that is a trained graph neural network (The examiner interprets “reference graph model” according to its broadest reasonable interpretation (BRI). The examiner notes that the applicant provides no meaningful definition of this term in their specification. The applicant only describes that a “reference graph model” is generated based on a trained graph neural network (see applicant’s specification at [0069]). Accordingly, the examiner interprets this term as encompassing a trained graph auto-encoder (GAE) as disclosed by Zhang below. A graph auto-encoder is a graph neural network trained to learn a lower dimension representation of a sensor graph (‘knowledge graph’). Thus, the resulting trained graph auto-encoder is a “reference graph model”.
Zhang discloses a graph auto-encoder (‘graph neural network’) uses a sensor graph (‘knowledge graph’) for anomaly detection, “To demonstrate the use of the sensor graph, we apply GNN to the sensor graph for the purpose of anomaly detection. In particular, we will apply the graph auto-encoder (GAE) [24]. As benchmark, we also consider conventional vector auto-encoder (VAE) [8] and principal component analysis (PCA) [15]. GAE compresses graph into hidden states (with lower dimension), and reconstructs it back. GAE is able to learn the hidden low dimensional structure in graphs directly” (P. 12, Sec. 4.1.3, ¶1).
Zhang discloses “The GAE used here consists of 3 layers … The target output of GAE is the original feature matrix, and backpropagation [8] is used to update the parameters” (P. 13, Sec. 4.1.3, First Paragraph).),
wherein the training comprises the graph neural network inferring weights capturing dynamics of the knowledge graph (Zhang discloses “a graph is G = (A, X), where A is the adjacency matrix, and X is the feature matrix … A(i, j) is the connection between two nodes i, j. …The nodes and edges information are both contained in the adjacency matrix A. Two nodes i, j are connected if A(i, j) ≠ 0. In this work, A will be symmetric, so the graph is undirected. In addition, A(i, j) can take any values in [0, 1], so it is a weighted graph” (P. 7, Sec. 2.6, ¶2-3).
Zhang discloses “The graph models both the features of time series (node feature matrix) and their connections (adjacency matrix), which makes it a good candidate as the knowledge representation between DC sensors” (P. 11, Sec. 4.1.1, Last Paragraph).
Zhang discloses “we define the following graph convolution operation: … where
A
∈
R
N
×
N
is the adjacency matrix,
X
k
∈
R
N
×
F
k
is the feature matrix at layer k. N is the number of nodes, and Fk is the number of features for each node.
A
^
is the normalized adjacency matrix, …
U
∈
R
N
×
N
,
W
∈
R
F
k
×
F
k
+
1
are parameters, and
B
∈
R
N
×
F
k
+
1
is the bias. Notice that graph adjacency matrix is kept fixed across GCN layers, while the feature matrix is transformed. The transformation of feature matrix depends on adjacency matrix, i.e., the topology of the graph. After convolution, a tanh activation function is be applied. The GAE used here consists of 3 layers. The first layer reduces the number of features per node from 4 to 3, and applies the tanh activation. The second layer reconstructs the number of features from 3 to 4, followed by the tanh transformation. The third layer is a rescaling layer, where only graph convolution is applied, without activation function. The target output of GAE is the original feature matrix, and backpropagation [8] is used to update the parameters” (P. 12-13, Sec. 4.1.3, ¶3-4).
Zhang discloses Equation 8 on P. 12 (reproduced below) describing a graph convolution operation for a graph auto-encoder. The graph convolution operation transforms the current feature matrix at each layer. A parameter (‘weight’) matrix W is aggregated with a normalized adjacency matrix and a feature matrix at layer k to transform the feature matrix for the next layer. A parameter matrix W is updated by backpropagation, therefore the parameters in the matrix are learnable parameters. Since an updated parameter matrix is used to transform a feature matrix, the learnable parameters in the parameter matrix capture “dynamics of the knowledge graph”.
PNG
media_image1.png
464
1220
media_image1.png
Greyscale
),
the weights being implemented in the reference graph model (Zhang discloses a graph auto-encoder’s (‘reference graph model’) parameters are updated (‘implemented’) using backpropagation, “The target output of GAE is the original feature matrix, and backpropagation [8] is used to update the parameters” (P. 12-13, Sec. 4.1.3, ¶3-4).)
and monitoring and detecting one or more anomalies for the industrial process based on the reference graph model (Zhang discloses “Subspace based method assumes that most data are “normal” and lie in some low dimensional subspace (linear or nonlinear). It tries to learn this subspace from data, then project data onto the subspace, and finally reconstruct the original input. “normal” data will have small reconstruction error, but “anomaly” data will have a higher reconstruction error. Therefore, one can differentiate “anomaly” from “normal” based on the reconstruction error” (P. 12, Sec. 4.1.3, ¶2).
Zhang further discloses “The anomaly score (“likelihood” of an anomaly) is based on the reconstruction error. Reconstruction error has the same dimension as the feature matrix, but we have a “sensor” level anomaly label. Therefore, we will aggregate the reconstruction error by taking the average across four features for each sensor. The anomaly score is defined to be this “sensor” level reconstruction error” (P. 14-15, Sec. 4.3.2).).
However, Zhang does not teach detecting an abnormal deviation in learned state-space parameters of the graph neural network relative to a reference state embedding, which is taught by Taslakian:
and monitoring and detecting one or more anomalies … based on the reference graph model (The Examiner interprets “reference graph model” according to its BRI. The examiner notes that the applicant provides no meaningful definition of this term in their specification. The applicant only describes that a “reference graph model” is generated based on a trained graph neural network (see applicant’s specification at [0069]). Accordingly, the Examiner interprets this term as encompassing a HideGNN model.
[0160] “The embodiments herein provide two self-supervised models referred to as Anomaly GNNs: HideGNN and MaskGNN. These methods are based on recreating the features of the nodes, either by hiding specific features, … The power of these methods comes from their ability to reproduce the original features of the nodes of the graph that corresponds to their expected values. By using these reproduced features or by comparing them to the original ones, anomalies arise as nodes with unexpected attributes.”
[0178-0179] “The HideGNN model uses a set of ƒ GNNs. Each of them will take F as input, where exactly one feature is hidden, and will use it to reproduce the hidden feature. … Thus, F(−i) has all the original features of the graph except the ith, which is removed. The ith loss function is then defined to be the L2 norm between the prediction of the network for the hidden feature and the real value of the hidden feature”
A HideGNN model is a graph neural network that is trained using the features of a set of GNNs with the exception of the ith features, which are removed, therefore the HideGNN model is a reference graph model since it was generated based on graph neural networks.),
wherein detecting the one or more anomalies comprises detecting an abnormal deviation in learned state-space parameters of the graph neural network relative to a reference state embedding (The Examiner interprets “state embedding” according to its BRI in view of the Applicant’s specification as encompassing a node embedding. This interpretation is consistent with the illustrative descriptions in the Applicants Specification at [0027] (see excerpt below).
Applicant’s written description at [0027] “the objective of GNN is to learn a state embedding, which contains the information pertaining to the neighborhood for each entity. The state embedding is a vector of the entity and can be used to produce an output such as the entity label.”
[0161] “the model tries to learn a general representation of the nodes by learning the characteristics of “normal” nodes. Any node in the graph that falls sufficiently far from this measure of normalcy will be classified as an anomaly.”
[0169] “The embeddings for a given node v takes into account the attributes of node v as well as its connectivity to neighboring nodes and the attributes of the neighboring nodes. The embeddings evolve into a form that can be used to predict when nodes are anomalous.”
[0170-0171] “The AGGREGATE function uses the previous embeddings of the node v and its neighbors u and combines them. … the AGGREGATE function is defined as the sum of the neighbors multiplied by a matrix of weights”
[0182] “The HideGNN model takes a graph in which the embeddings have been determined, … the first attribute of each node is hidden, and then the second and third are used to derive (guess) an embedding for the hidden attribute. This hiding and projecting process continues for all attributes”
[0183] “An anomaly is flagged when the embeddings derived from the hiding process deviates from the “known” embeddings by more than a pre-determined amount. A mean squared distance calculation or Euclidean distance calculation may be used to determine the deviation. The pre-determined amount may be, for example, based on a distribution of distances calculated over the “known” embeddings. Thus, in some embodiments, if the deviation is more than two or three standard deviations from the mean of this distribution, the node may be determined to be anomalous.”
A node embedding is generated by aggregating a given node’s previous embedding with the sum of the given node’s neighbors multiplied by a matrix of weights, therefore an embedding contains information (attributes) about the given node’s neighborhood (and therefore is a state embedding). The HideGNN model produces node embeddings (‘reference state embeddings’) during its hiding process. A node embedding is generated based on a weight matrix and weights are updated during GNN training (see [0149]), therefore node embeddings are learned state-space parameters of a graph neural network.
The distance between an embedding derived from the HideGNN model (reference state embedding) and a known node embedding is used to classify a node as anomalous (therefore detecting an abnormal deviation in learned state-space parameters of the graph neural network relative to a reference state embedding).).
Before the effective filing date of the claimed invention, it would have been obvious to combine the anomaly detection method of Zhang with the technique disclosed by Taslakian to detect anomalies based on node embedding distance. By detecting anomalies based on node embedding distance, standard deviations of distance distributions can be used to classify nodes as anomalous, thereby increasing anomaly detection accuracy as standard deviations can be used as a baseline to measure how far node embeddings deviate from expected norms.
With respect to claims 7, 14 and 20, the combined anomaly detection method of Zhang/Taslakian teaches:
the method of claim 1, further including scoring each of the one or more anomalies based on a determination of a degree of error between the knowledge graph and the reference graph model (Zhang discloses “GAE compresses graph into hidden states (with lower dimension), and reconstructs it back. GAE is able to learn the hidden low dimensional structure in graphs directly. VAE learns the nonlinear low dimensional subspace on vector inputs. PCA works by finding a linear subspace (of vectors) that best explains the variance in data. Both these three methods fall into the category of subspace-based methods for anomaly detection [5]. Subspace based method assumes that most data are “normal” and lie in some low dimensional subspace (linear or nonlinear). It tries to learn this subspace from data, then project data onto the subspace, and finally reconstruct the original input. “normal” data will have small reconstruction error, but “anomaly” data will have a higher reconstruction error. Therefore, one can differentiate “anomaly” from “normal” based on the reconstruction error” (P. 12, Sec. 4.1.3, ¶1-2).
Zhang further discloses “The anomaly score (“likelihood” of an anomaly) is based on the reconstruction error. Reconstruction error has the same dimension as the feature matrix, but we have a “sensor” level anomaly label. Therefore, we will aggregate the reconstruction error by taking the average across four features for each sensor. The anomaly score is defined to be this “sensor” level reconstruction error. Based on the anomaly score, one can obtain a precision (F1 score)-recall curve by adjusting the threshold of anomaly prediction” (P. 14, Sec. 4.3.2, ¶1).
A graph auto-encoder (‘reference graph model’) transforms a sensor graph (‘knowledge graph’) into a low dimensional representation. The graph auto-encoder then tries to reconstruct the low dimensional representation of the sensor graph back to its original representation. Anomalies are detected based on a reconstruction error since “normal” data will have a small reconstruction error, but anomalous data will have a higher reconstruction error. Therefore, anomalies are detected based on ‘a degree of error’ between the representation of a knowledge graph and the representation produced by a reference graph model. An anomaly score based on a reconstruction error is then generated to determine the likelihood of an anomaly.).
With respect to claim 8, the rejection of claim 1 is incorporated. The difference in scope being
A computer system comprising (Zhang discloses “To demonstrate the use of the DC sensor graph for anomaly detection, we compare the performance of graph neural network (GNN) and existing standard methods on synthetic anomaly data” (P. 1, Abstract).):
one or more processors (A computer is implied by comparing the performance of a GNN, which further implies one or more processors.);
one or more computer readable storage media (A computer is implied by comparing the performance of a GNN, which further implies one or more computer readable storage media.);
and executable instructions stored on the one or more computer readable storage media that when executed cause the one or more processors to perform operations comprising (A computer is implied by comparing the performance of a GNN, which further implies a computer readable storage media storing instructions to be executed by a processor.).
With respect to claim 15, the rejection of claim 1 is incorporated. The difference in scope being
A computer program product comprising (Zhang discloses “To demonstrate the use of the DC sensor graph for anomaly detection, we compare the performance of graph neural network (GNN) and existing standard methods on synthetic anomaly data” (P. 1, Abstract).):
one or more computer readable storage media (A computer is implied by comparing the performance of a GNN, which further implies one or more computer readable storage media.),
and program instructions stored on the one or more computer readable storage media to perform operations comprising (A computer is implied by comparing the performance of a GNN, which further implies a computer readable storage media storing instructions to be executed.).
With respect to claims 29 and 31, the combined anomaly detection method of Zhang/Taslakian teaches the method of claim 1, wherein the reference graph model determines one or more classifications of the one or more entities of the knowledge graph (Zhang discloses sensors (‘entities’) are labeled (‘classified’) as anomalies based on a reconstruction error, “The anomaly score (“likelihood” of an anomaly) is based on the reconstruction error. Reconstruction error has the same dimension as the feature matrix, but we have a “sensor” level anomaly label. Therefore, we will aggregate the reconstruction error by taking the average across four features for each sensor. The anomaly score is defined to be this “sensor” level reconstruction error” (P. 14, Sec. 4.3.2, ¶1).
Zhang discloses a graph auto-encoder (‘reference graph model’) is used to determine a reconstruction error, see (P. 12, Sec. 4.1.3, ¶1-2).).
The following are the references relied upon in the rejections below:
Creed (US 20210081717 A1)
Claims 4, 6, 11, 13, 17, 19, 22, 27, 30 and 32 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Taslakian, further in view of Creed.
With respect to claims 4, 11, and 17, the combined anomaly detection method of Zhang/Taslakian teaches the method of claim 1, however the combination does not teach configuring one or more graph subsets, which is taught by Creed:
further including configuring one or more graph subsets identified from the knowledge graph using the graph neural network (Creed discloses a filtered entity-entity graph (‘knowledge graph’) as a ‘graph subset’, “Creed discloses “the trained attention weights of the GNN model may be output for use as the application demands. The trained attention weights that are output may be used to filter the entity-entity relationship graph by weighting each edge with the corresponding trained attention weight. Filtering the entity-entity graph may further include removing a relationship edge from the entity-entity graph if the corresponding attention weight is below and/or equal to an attention threshold. The attention threshold may be predetermined or adjusted depending on the amount of perceived noise in the entity-entity graph. This may be used to improve the entity-entity relationship graph, which may in turn be used to further train another GNN model or even update the present trained GNN model. The GNN model may be further updated based on the filtered entity-entity relationship graph” [0109].).
Creed teaches filtering an entity-entity graph using trained attention weights of a graph neural network is a known method in the art. Before the effective filing date of the claimed invention, it would have been obvious to modify the combined anomaly detection method of Zhang/Taslakian with the graph filtering technique of Creed to remove noisy edges from a knowledge graph. By removing noisy edges from a knowledge graph, only the edges that capture relevant and important information can remain, thus saving storage space and increasing interpretability of the knowledge graph.
With respect to claims 6, 13 and 19, the combined anomaly detection method of Zhang/Taslakian teaches the method of claim 1, however the combination does not teach generating one or more vector functions for each entity in a knowledge graph, which is taught by Creed:
further including generating one or more vector functions for each one of the entities in the knowledge graph (Creed discloses “conventional GCNNs may be used to generate an embedding of graph 102 by using convolutional graph neural network operations for embedding each entity node as an N-dimensional vector using relationship scoring. The convolutional operation is briefly outlined below and described in terms of building or generating layers of embeddings (or convolutions of embeddings of the entity nodes). Essentially, each embedding layer builds upon the previous embedding layer. For example, an embedding for disease entity node D 103a for a layer l+1 may be represented, by way of example only but is not limited to, an N-dimensional vector … where σ(⋅) as an activation function (e.g. ReLu),
W
r
l
is an N×N dimensional adjacency matrix for layer l representing relationships associated with relation index rϵR between entity node D and protein node P in graph G 102, R being the set of relations, … Each of the embeddings for the other entity nodes P and C 103b-103g may be calculated in a similar fashion” [0097].
See [0097] describing an N-dimensional vector that is generated for each entity node in an entity graph (‘knowledge graph’). Each N-dimensional vector is generated by using a convolutional operation (‘vector function’).).
Creed teaches using convolutional operations to generate N-dimensional vectors for each entity node in an entity graph is a known method in the art. Before the effective filing date of the claimed invention, it would have been obvious to modify the combined anomaly detection method of Zhang/Taslakian with the technique disclosed by Creed to generate N-dimensional vectors for each entity node. By using convolutional operations to generate N-dimensional vectors, entities and their relationships can be represented in a lower dimensional space, thereby allowing for more efficient computational processing and modeling.
With respect to claims 22 and 27, the combined anomaly detection method of Zhang/Taslakian teaches the method of claim 1, however, the combination does not teach incorporating data extracted from an external knowledge base into a knowledge graph, which is taught by Creed:
wherein data extracted from an external knowledge base is further incorporated into the knowledge graph (Creed discloses “Large-scale databases of facts can be represented as a entity-entity knowledge graph, which is a graph structure describing relationships between entity nodes, that is used to form the basis of a prediction or increase prediction quality, and offer facts in the universal form of the graph triple e=(s,r,o) that relates a subject entity node s through a relation edge r to an object entity node o of the knowledge graph” [0002].
Creed further discloses “The generated GNN model using the modified GCNN technique is capable of indirectly or explicitly filtering noisy or incorrect connections of a entity-entity knowledge graph, which may be generated from a database of facts or information such as, by way of example only but not limited to, a large-scale database, a corpus of information, literature, text and the like” [0082].).
Creed teaches representing a large-scale database of facts as an entity-entity knowledge graph is a known method in the art. Before the effective filing date of the claimed invention, it would have been obvious to modify the combined anomaly detection method of Zhang/Taslakian with the technique disclosed by Creed to generate a knowledge graph from a large-scale database. By generating a knowledge graph from a large-scale database, a knowledge graph can be constructed with real-world data and context, providing a more accurate representation of entities and their real-world relationships.
With respect to claims 30 and 32, the combined anomaly detection method of Zhang/Taslakian teaches the method of claim 1, however the combination does not teach predicting one or more links between entities of a knowledge graph, which is taught by Creed:
wherein the reference graph model predicts one or more links between the one or more entities of the knowledge graph (The examiner interprets “reference graph model” according to its BRI. The examiner notes that the applicant provides no meaningful definition of this term in their specification. The applicant only describes that a “reference graph model” is generated based on a trained graph neural network (see applicant’s specification at [0069]). Accordingly, the examiner interprets this term as encompassing an updated graph neural network as disclosed by Creed below.
Creed discloses a GNN is trained to learn an entity-entity knowledge graph, “generating and training a robust graph neural network (GNN) model using attention weights and based on data representative of at least a portion of an entity-entity graph dataset generated from, by way of example only but not limited to, a dataset of facts and/or entity relations/relationships and the like” [0074].
Creed discloses “the trained attention weights of the GNN model may be output for use as the application demands. The trained attention weights that are output may be used to filter the entity-entity relationship graph by weighting each edge with the corresponding trained attention weight. Filtering the entity-entity graph may further include removing a relationship edge from the entity-entity graph if the corresponding attention weight is below and/or equal to an attention threshold. The attention threshold may be predetermined or adjusted depending on the amount of perceived noise in the entity-entity graph. This may be used to improve the entity-entity relationship graph, which may in turn be used to further train another GNN model or even update the present trained GNN model. The GNN model may be further updated based on the filtered entity-entity relationship graph. The updated GNN model may be used to predict link relationships between entities in the filtered entity-entity relationship graph by inputting to the updated GNN model representations of two or more entities and/or one or more relationships; and receiving an indication of the likelihood of a link relationship existing between the two or more entities” [0109].
A trained GNN model’s attention weights are used to filter an entity-entity knowledge graph. The filtered entity-entity knowledge graph can then be used to update the trained GNN model, resulting in an updated GNN model (‘reference graph model’). The updated GNN may then be used to predict link relationships between entities of a filtered knowledge graph.).
Creed teaches predicting link relationships between entities of a knowledge graph is a known method in the art. Before the effective filing date of the claimed invention, it would have been obvious to modify the combined anomaly detection method of Zhang/Taslakian with the link prediction technique disclosed by Creed to predict link relationships between entities. By predicting link relationships between entities, incomplete knowledge graphs can be completed by identifying the most likely missing links to form new facts, thereby enabling better data analysis.
The following are the references relied upon in the rejections below:
Kumar Jaya Kumar (US 20230050889 A1)
Claims 21 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Taslakian, further in view of Creed and Kumar Jaya Kumar.
With respect to claims 21 and 26, the combined anomaly detection method of Zhang/Taslakian/Creed teaches the method of claim 4, however, the combination does not teach inputting graph subsets into a pipeline configuration to provide context for a contextualized diagnosis of anomalies, which is taught by Kumar Jaya Kumar:
further comprising inputting at least one of the one or more graph subsets into a pipeline configuration such that, in response, the pipeline configuration provides context for a contextualized diagnosis of the one or more anomalies (The examiner interprets “pipeline configuration” according to its BRI in view of applicant’s specification at [00102-00103]. Accordingly, the examiner interprets this term as encompassing a series of data processing and modeling steps that automate anomaly diagnosis as disclosed by the root cause analysis system of Kumar Jaya Kumar below.
Kumar Jaya Kumar discloses a root cause analysis (RCA) system (‘pipeline configuration’), “disclosure provides an improved and efficient method and an RCA system that dynamically performs knowledge graph and sub-graph clusters based RCA. The solution provided by the present disclosure has the automation capability to learn (technical and/or business) domain and generate the causes for failure by using the unsupervised machine learning technique. The domain learning is represented in terms of knowledge graph and sub-graph clusters. The present disclosure processes received input content to extract a set of features in order to generate a knowledge graph and thereafter, a set of sub-graphs from the knowledge graph. The knowledge graph and sub-graphs help to build the domain knowledge. Using the generated knowledge graph and the sub-graphs, the graph data structures are extracted. The extracted graph data structures from the knowledge graph and the sub-graphs are processed to generate sub-graph cluster(s) and corresponding probabilistic graphical model. The probabilistic graphical model helps to determine the core problems that led to the root cause analysis and act as a core root cause classifier. Once the core root cause is determined, a probabilistic graphical model is built for each cluster that is available in the core root cause classifier. Thereafter, whenever, an issue content containing an issue is received, the present disclosure determines a root cause for the issue using the knowledge graph, the root cause model and information related to the at least one sub-graph cluster and corresponding probabilistic graphical model for each of the sub-graph cluster” [0025].).
Kumar Jaya Kumar teaches processing subgraphs to determine the root cause of an issue (‘anomaly’) is a known method in the art. Before the effective filing date of the claimed invention, it would have been obvious to modify the combined anomaly detection method of Zhang/Taslakian/Creed with the root cause analysis system of Kumar Jaya Kumar to use subgraphs to identify the root cause of anomalies. By using subgraphs to identify the root cause of anomalies, attention can be focused on analyzing smaller, relevant subgraphs, thereby capturing local patterns and improving the interpretability of the anomaly detection process.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PEDRO J MORALES whose telephone number is (571)272-6106. The examiner can normally be reached 8:30 AM - 6:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, MIRANDA M HUANG can be reached at (571)270-7092. 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.
/PEDRO J MORALES/Examiner, Art Unit 2124
/VINCENT GONZALES/Primary Examiner, Art Unit 2124
Prosecution Insights