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 .
Response to Arguments
Applicant's arguments filed 04/24/2026 have been fully considered but they are not fully persuasive.
Regarding the 101 rejections, applicant’s arguments and amendments to the independent claims are persuasive and overcome the previous 101 rejections. Specifically, applicant’s amended limitations of generating dependency graph objects from the windowed subsequences utilizing a sparse graph recovery model; determining one or more segmentation timestamps when one or more segment changes in the multivariate time series data occurred based on comparing the dependency graph objects utilizing a similarity model, wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps; and generating a segmented multivariate time series by segmenting the multivariate time series data based on the one or more segmentation timestamps provides a technical improvement because using a sparse recovery model to generate dependency graph and identifying segmentation timestamps by comparing the dependency graphs in a similarity model improves the segmentation process of multivariate time series. See pg. 13 of “Remarks”: “The Specification identifies the technical problem that conventional multivariate time-series segmentation systems are inadequate, inefficient, and inflexible. Furthermore, conventional approaches typically operate at quadratic complexity (e.g., O(N2)). See id., [0001], [0016]- [0017]. The disclosed solution is equally direct and involves generating graph objects from short intervals of the multivariate time series using a graph recovery model, and determining segmentation timestamps from those graph objects using a similarity model. See id., [0018], [0040]-[0042]. More specifically, the Specification explains the technical benefit of that architecture. Paragraph [0019] states that the time series segmentation system achieves segmented multivariate time series in O(N) operations by converting the input time series into a temporal dependency graph-object sequence using a graph recovery model, which is a significant improvement over conventional systems that require O(N2) operations at best. Paragraph [0020] explains that the use of dependency graphs makes the system flexible, scalable, and cross- domain. Paragraph [0021] explains that sparse graph recovery models can recover graph objects for multiple windowed subsequences in a single batch using shared parameters. Paragraph [0022] explains that the graph objects provide interpretability beyond conventional sparse-matrix approaches. These benefits flow from the very architecture now recited in the amended claims. To illustrate, the currently amended independent claims recite generating dependency graph objects using a sparse graph recovery model and using a similarity model that compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change and determine segmentation timestamps. In other words, the claims now recite the technical mechanism that the Specification identifies as producing the improved segmentation process. The claims, therefore, do more than merely invoke a computer to apply the alleged abstract idea. They recite a specific technological arrangement for converting multivariate time-series data into a segmented multivariate time series. See id., [0018]-[0022], [0029], [0031]- [0034], [0040]-[0042].” Applicant’s amendments and corresponding arguments that the claimed invention provides a technical improvement to the field of multivariate time series segmentation are persuasive. Therefore, the 101 rejections are withdrawn.
Regarding the 103 rejections, applicant's arguments filed with respect to the prior art rejections have been fully considered but they are moot. Applicant has amended the claims to recite new combinations of limitations. Applicant's arguments are directed at the amendment. Please see below for new grounds of rejection, necessitated by Amendment.
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.
Claims 1-3, 6, 8-9, 11, 17, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, et al., US Pre-Grant Publication 2020/0372024A1 (“Kulkarni”) in view of Shrivastava, et al., Non-Patent Literature “GLAD: LEARNING SPARSE GRAPH RECOVERY” (“Shrivastava”) and further in view of Sulem, et al., Non-Patent Literature “Graph similarity learning for change-point detection in dynamic networks” (“Sulem”).
Regarding claim 1, Kulkarni discloses:
A computer-implemented method for generating segmented multivariate time series data comprising: (Kulkarni, ⁋21, “The multi-scale segmentation module 104 receives one or more time-series datasets 102. The datasets 102 may be, for example, multivariate time-series datasets (MTD) which include time-series data for different assets (for example, IoT sensor data) [A computer-implemented method for generating segmented multivariate time series data comprising:].”).
grouping portions of multivariate time series data by a window size to generate windowed subsequences of the multivariate time series data; (Kulkarni, ⁋22, “The multi-scale segmentation module 104 creates segments of the time series datasets 102 of different sizes [grouping portions of multivariate time series data by a window size]. FIG. 2 depicts a non-limiting example of multi-scale segmentation 200 in accordance with one or more example embodiments. As shown in FIG. 2, it is assumed the total data duration for the time series of each asset's data is n, so that the sizes may be, for example, n, n/2, n/4, wherein each size represents a different time scale [to generate windowed subsequences of the multivariate time series data;].”).
generating dependency graph objects from the windowed subsequences…; (Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 constructs a segment similarity graph for each scale using the distance matrix [generating dependency graph objects from the windowed subsequences…;].”).
determining one or more segmentation timestamps when one or more segment changes in the multivariate time series data occurred based on comparing the dependency graph objects utilizing a similarity model…; (Kulkarni, ⁋31, “FIG. 3B is a segment graph 350 with segments from four different assets, namely, assets 1-4. The segment graph in FIG. 3B shows anomalous segment closeness between the assets. In particular, segments of asset 3 are closely grouped with a segment of asset 2, and a segment of asset 4 is closely grouped with a segment of asset 2; grouping the anomalous segments closer together based on closeness is interpreted as determining an anomalous timesteps or points in the dataset (i.e. determining one or more segmentation timestamps when one or more segment changes in the multivariate time series data occurred based on comparing the dependency graph objects).”, and Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 [utilizing a similarity model…;] constructs a segment similarity graph for each scale using the distance matrix.”).
and generating a segmented multivariate time series by segmenting the multivariate time series data based on the one or more segmentation timestamps. (Kulkarni, ⁋31, “FIG. 3B is a segment graph 350 with segments from four different assets, namely, assets 1-4. The segment graph in FIG. 3B shows anomalous segment closeness between the assets. In particular, segments of asset 3 are closely grouped with a segment of asset 2, and a segment of asset 4 is closely grouped with a segment of asset 2; grouping segments closer together based on anomalous closeness is interpreted as generating a segmented multivariate time series using segmentation timesteps, or anomalous points in the data (i.e. and generating a segmented multivariate time series by segmenting the multivariate time series data based on the one or more segmentation timestamps.).”).
While Kulkarni teaches a system that segments multivariate time series data using graph objects, Kulkarni does not explicitly teach:
…utilizing a sparse graph recovery model…
…wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps…
Shrivastava teaches …utilizing a sparse graph recovery model… (Shrivastava, abstract, “We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data […utilizing a sparse graph recovery model…].”).
Kulkarni and Shrivastava are both in the same field of endeavor (i.e. graph analysis). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni and Shrivastava to teach the above limitation(s). The motivation for doing so is that recovering sparse conditional independence graphs improves understanding of relationships between different assets (cf. Shrivastava, pg. 1, “Recovering sparse conditional independence graphs from data is a fundamental problem in high dimensional statistics and time series analysis, and it has found applications in diverse areas. In computational biology, a sparse graph structure between gene expression data may be used to understand gene regulatory networks; in finance, a sparse graph structure between financial time series may be used to understand the relationship between different financial assets.”).
While the combination teaches a system for segmenting multivariate time series data, the combination does not explicitly teach:
…wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps…
Sulem teaches …wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps… (Sulem, pg. 2, “For multivariate time series, change-point detection is a task that has been widely studied in various settings (e.g., nonparametric [9], high-dimensional [10] or online [11]). The equivalent task for dynamic networks is often denoted network change-point detection (NCPD) and has recently become a popular problem with numerous successful ap plications in finance [7], neuroscience [12] or transport networks [13].”, and Sulem, pg. 5, “In our NCPD method [to identify a change associated with a segment change to determine the one or more segmentation timestamps…], we propose a s-GNN architecture summarized in Figure 1, for learning a similarity score s(Gt1 ,Gt2 ) in [0,1] on the space of graph snapshots (G1,G2,...,Gt,...) from the dynamic network [compares corresponding correlations of the dependency graph objects]. For this purpose, we design a similarity module […wherein the similarity model] for comparing the node-level embeddings output by a generic graph encoder (e.g., a graph convolutional network [40], a graph attention network [57], a GraphSage network [58] or a graph isomorphism network (GIN) [59]).”).
Kulkarni, in view of Shrivastava, and Sulem are both in the same field of endeavor (i.e. time series segmentation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava, and Sulem to teach the above limitation(s). The motivation for doing so is that using a siamese graph neural network as a similarity model improves the robustness of the similarity analysis between graph structures (cf. Sulem, pg. 2, “We propose a graph similarity learning model based on a siamese graph neural network able to handle any available node attributes, and demonstrate how it can be leveraged for the online NCPD problem with an adequate training procedure. In particular, our learnt similarity function is sensitive to both local and global displacements in the graph structure, and can effectively be employed in the context of change-point (and anomaly) detection in temporal networks.”).
Regarding claim 2, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1. Kulkarni further teaches further comprising: comparing a first dependency graph object to a second dependency graph object utilizing the similarity model to determine that a difference between the first dependency graph object and the second dependency graph object satisfies a difference threshold; and determining a first segmentation timestamp based on a segmentation timestamp of the first dependency graph object. (Sulem, pg. 5, “In our NCPD method, we propose a s-GNN architecture summarized in Figure 1, for learning a similarity score s(Gt1 ,Gt2 ) [further comprising: comparing a first dependency graph object to a second dependency graph object utilizing the similarity model] in [0,1] on the space of graph snapshots (G1,G2,...,Gt,...) from the dynamic network…Therefore, for a pair of graphs (Gt1 ,Gt2 ), this siamese encoder module computes a pairs of graph embeddings, (H1, H2) := (HJ(Gt1 ), HJ(Gt2 )), and the vectors (H1)i: and (H2)i: correspond to the representations of the node i respectively in Gt1 and Gt2 . Intuitively, a large distance between these two embeddings can indicate that node i plays distinct structural roles in Gt1 and Gt2. Then, the pair of embeddings (H1, H2) is processed by a similarity module, which first computes a vector of Euclidean distance between the nodes’ embeddings [to determine that a difference between the first dependency graph object and the second dependency graph object satisfies a difference threshold;], and secondly, applies a Sort-k pooling operation [60] to select its k largest entries, i.e., P =(fr1 ,..., frk ), fi = (H1)i: − (H2)i: 2 ∈ 0, 1 i n, where r1,...,rk correspond to the indices of the (sorted) k largest elements of {fi}i∈[n]…so that the final output of the similarity module (and the s-GNN), s(Gt1 ,Gt2 ) ∈ [0,1], a non-negative similarity score between the two input graphs.”, and Sulem, pg. 4, “Our NCPD method consists of using the statistic Zt(s,L) and the detection rule (2), together with a data-driven graph similarity function s(·, ·) learnt by a s-GNN model, which we describe in the next section [and determining a first segmentation timestamp based on a segmentation timestamp of the first dependency graph object.].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Sulem with the teachings of Kulkarni and Shrivastava for the same reasons disclosed in claim 1.
Regarding claim 3, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1. Kulkarni further teaches wherein generating the dependency graph objects from the windowed subsequences includes generating a visual graph of nodes and edges, where the edges indicate a positive or a negative partial correlation between connected nodes. (Kulkarni, ⁋25, “FIG. 3A shows an example of a segment graph 300 [wherein generating the dependency graph objects from the windowed subsequences includes generating a visual graph of nodes and edges,] wherein the vertices represent data coming from all assets for a given-time scale, and the edges represent the distance between the distributions of the data represented by the vertices [where the edges indicate a positive or a negative partial correlation between connected nodes.].”).
Regarding claim 6, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1. Shrivastava further teaches wherein generating the dependency graph objects from the windowed subsequences includes utilizing a conditional independence sparse graph recovery model that generates conditional independence graph objects that exhibit partial correlation between variables. (Shrivastava, pg. 1, “Recovering sparse conditional independence graphs from data is a fundamental problem in high dimensional statistics and time series analysis [wherein generating the dependency graph objects from the windowed subsequences includes utilizing a conditional independence sparse graph recovery model]”, and Shrivastava, pg. 2, “Given m observations of ad-dimensional multivariate Gaussian random variable X = [X1,...,Xd]⊤, the sparse graph recovery problem aims to estimate its covariance matrix Σ∗ and precision matrix Θ∗ =(Σ∗)−1. The ij-th component of Θ∗ is zero if and only if Xi and Xj are conditionally independent given the other variables {Xk}k̸=i,j [that generates conditional independence graph objects that exhibit partial correlation between variables.].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Shrivastava with the teachings of Kulkarni and Sulem for the same reasons disclosed in claim 1.
Regarding claim 8, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1. Sulem further teaches wherein the similarity model includes an allocation algorithm that determines the one or more segmentation timestamps based on determining a first order distance and a second order distance from the dependency graph objects. (Sulem, pg. 5, “In our NCPD method, we propose a s-GNN architecture summarized in Figure 1, for learning a similarity score s(Gt1 ,Gt2 ) [wherein the similarity model includes an allocation algorithm that determines the one or more segmentation timestamps] in [0,1] on the space of graph snapshots (G1,G2,...,Gt,...) from the dynamic network…Therefore, for a pair of graphs (Gt1 ,Gt2 ), this siamese encoder module computes a pairs of graph embeddings, (H1, H2) := (HJ(Gt1 ), HJ(Gt2 )), and the vectors (H1)i: and (H2)i: correspond to the representations of the node i respectively in Gt1 and Gt2 . Intuitively, a large distance between these two embeddings can indicate that node i plays distinct structural roles in Gt1 and Gt2. Then, the pair of embeddings (H1, H2) is processed by a similarity module, which first computes a vector of Euclidean distance between the nodes’ embeddings, and secondly, applies a Sort-k pooling operation [60] to select its k largest entries, i.e., P =(fr1 ,..., frk ), fi = ||(H1)i: − (H2) i:|| 2 [based on determining a first order distance and a second order distance from the dependency graph objects.] ∈ 0, 1 i n, where r1,...,rk correspond to the indices of the (sorted) k largest elements of {fi}i∈[n].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Sulem with the teachings of Kulkarni and Shrivastava for the same reasons disclosed in claim 1.
Regarding claim 9, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 8. Sulem further teaches:
wherein: the first order distance captures a distance between consecutive dependency graph objects; (Sulem, pg. 5, “Then, the pair of embeddings (H1, H2) is processed by a similarity module, which first computes a vector of Euclidean distance between the nodes’ embeddings, and secondly, applies a Sort-k pooling operation [60] to select its k largest entries, i.e., P =(fr1 ,..., frk ), fi = ||(H1)i: − (H2) [wherein: the first order distance captures a distance between consecutive dependency graph objects;] i:|| 2 ∈ 0, 1 i n, where r1,...,rk correspond to the indices of the (sorted) k largest elements of {fi}i∈[n].”).
and the second order distance generates absolute values based on the first order distance. (Sulem, pg. 5, “Then, the pair of embeddings (H1, H2) is processed by a similarity module, which first computes a vector of Euclidean distance between the nodes’ embeddings, and secondly, applies a Sort-k pooling operation [60] to select its k largest entries, i.e., P =(fr1 ,..., frk ), fi = ||(H1)i: − (H2) i:|| 2 [and the second order distance generates absolute values based on the first order distance.] ∈ 0, 1 i n, where r1,...,rk correspond to the indices of the (sorted) k largest elements of {fi}i∈[n].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Sulem with the teachings of Kulkarni and Shrivastava for the same reasons disclosed in claim 8.
Regarding claim 11, Kulkarni discloses:
A system comprising: multivariate time series data; (Kulkarni, ⁋21, “The multi-scale segmentation module 104 receives one or more time-series datasets 102. The datasets 102 may be, for example, multivariate time-series datasets (MTD) which include time-series data for different assets (for example, IoT sensor data) [A system comprising: multivariate time series data;].”).
a similarity model that determines differences between two or more dependency graph objects; (Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 constructs a segment similarity graph for each scale using the distance matrix [a similarity model that determines differences between two or more dependency graph objects;].”).
a processor; and a computer memory comprising instructions that, when executed by the processor, cause the system to carry out operations comprising: (Kulkarni, ⁋5, “a system including a memory and at least one processor that is coupled to the memory and configured to perform noted method steps [a processor; and a computer memory comprising instructions that, when executed by the processor, cause the system to carry out operations comprising:]”).
generating windowed subsequences of the multivariate time series data by grouping portions of the multivariate time series data by a time-based window size; (Kulkarni, ⁋22, “The multi-scale segmentation module 104 creates segments of the time series datasets 102 of different sizes [grouping portions of the multivariate time series data by a time-based window size;]. FIG. 2 depicts a non-limiting example of multi-scale segmentation 200 in accordance with one or more example embodiments. As shown in FIG. 2, it is assumed the total data duration for the time series of each asset's data is n, so that the sizes may be, for example, n, n/2, n/4, wherein each size represents a different time scale [generating windowed subsequences of the multivariate time series data].”).
generating dependency graph objects from the windowed subsequences…; (Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 constructs a segment similarity graph for each scale using the distance matrix [generating dependency graph objects from the windowed subsequences…;].”).
determining one or more segmentation timestamps based on the dependency graph objects utilizing the similarity model…; (Kulkarni, ⁋31, “FIG. 3B is a segment graph 350 with segments from four different assets, namely, assets 1-4. The segment graph in FIG. 3B shows anomalous segment closeness between the assets. In particular, segments of asset 3 are closely grouped with a segment of asset 2, and a segment of asset 4 is closely grouped with a segment of asset 2; grouping the anomalous segments closer together based on closeness is interpreted as determining an anomalous timesteps or points in the dataset (i.e. determining one or more segmentation timestamps based on the dependency graph objects utilizing the similarity model…;).”, and Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 [utilizing a similarity model;] constructs a segment similarity graph for each scale using the distance matrix.”).
and generating a segmented multivariate time series by segmenting the multivariate time series data based on one or more segmentation timestamps. (Kulkarni, ⁋31, “FIG. 3B is a segment graph 350 with segments from four different assets, namely, assets 1-4. The segment graph in FIG. 3B shows anomalous segment closeness between the assets. In particular, segments of asset 3 are closely grouped with a segment of asset 2, and a segment of asset 4 is closely grouped with a segment of asset 2; grouping segments closer together based on anomalous closeness is interpreted as generating a segmented multivariate time series using segmentation timesteps, or anomalous points in the data (i.e. and generating a segmented multivariate time series by segmenting the multivariate time series data based on one or more segmentation timestamps.).”).
While Kulkarni teaches a system that segments multivariate time series data using graph objects, Kulkarni does not explicitly teach:
a sparse graph recovery model that generates dependency graph objects from portions of…time series data;
…utilizing a sparse graph recovery model…
…wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps…
Shrivastava teaches:
a sparse graph recovery model that generates dependency graph objects from portions of…time series data; (Shrivastava, abstract, “We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data [a sparse graph recovery model that generates dependency graph objects].”, and Shrivastava, pg. 1, “Recovering sparse conditional independence graphs from data is a fundamental problem in high dimensional statistics and time series analysis [from portions of…time series data;]”).
…utilizing a sparse graph recovery model… (Shrivastava, abstract, “We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data […utilizing a sparse graph recovery model…].”).
Kulkarni and Shrivastava are both in the same field of endeavor (i.e. graph analysis). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni and Shrivastava to teach the above limitation(s). The motivation for doing so is that recovering sparse conditional independence graphs improves understanding of relationships between different assets (cf. Shrivastava, pg. 1, “Recovering sparse conditional independence graphs from data is a fundamental problem in high dimensional statistics and time series analysis, and it has found applications in diverse areas. In computational biology, a sparse graph structure between gene expression data may be used to understand gene regulatory networks; in finance, a sparse graph structure between financial time series may be used to understand the relationship between different financial assets.”).
While the combination teaches a system for segmenting multivariate time series data, the combination does not explicitly teach:
…wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps…
Sulem teaches …wherein the similarity model compares corresponding correlations of the dependency graph objects to identify a change associated with a segment change to determine the one or more segmentation timestamps… (Sulem, pg. 2, “For multivariate time series, change-point detection is a task that has been widely studied in various settings (e.g., nonparametric [9], high-dimensional [10] or online [11]). The equivalent task for dynamic networks is often denoted network change-point detection (NCPD) and has recently become a popular problem with numerous successful ap plications in finance [7], neuroscience [12] or transport networks [13].”, and Sulem, pg. 5, “In our NCPD method [to identify a change associated with a segment change to determine the one or more segmentation timestamps…], we propose a s-GNN architecture summarized in Figure 1, for learning a similarity score s(Gt1 ,Gt2 ) in [0,1] on the space of graph snapshots (G1,G2,...,Gt,...) from the dynamic network [compares corresponding correlations of the dependency graph objects]. For this purpose, we design a similarity module […wherein the similarity model] for comparing the node-level embeddings output by a generic graph encoder (e.g., a graph convolutional network [40], a graph attention network [57], a GraphSage network [58] or a graph isomorphism network (GIN) [59]).”).
Kulkarni, in view of Shrivastava, and Sulem are both in the same field of endeavor (i.e. time series segmentation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava, and Sulem to teach the above limitation(s). The motivation for doing so is that using a siamese graph neural network as a similarity model improves the robustness of the similarity analysis between graph structures (cf. Sulem, pg. 2, “We propose a graph similarity learning model based on a siamese graph neural network able to handle any available node attributes, and demonstrate how it can be leveraged for the online NCPD problem with an adequate training procedure. In particular, our learnt similarity function is sensitive to both local and global displacements in the graph structure, and can effectively be employed in the context of change-point (and anomaly) detection in temporal networks.”).
Regarding claim 17, Kulkarni discloses:
A computer-implemented method for generating segmented multivariate time series data comprising: (Kulkarni, ⁋21, “The multi-scale segmentation module 104 receives one or more time-series datasets 102. The datasets 102 may be, for example, multivariate time-series datasets (MTD) which include time-series data for different assets (for example, IoT sensor data) [A computer-implemented method for generating segmented multivariate time series data comprising:].”).
generating a first windowed subsequence and a second windowed subsequence from multivariate time series data based on a window size; (Kulkarni, ⁋22, “The multi-scale segmentation module 104 creates segments of the time series datasets 102 of different sizes [based on a window size;]. FIG. 2 depicts a non-limiting example of multi-scale segmentation 200 in accordance with one or more example embodiments. As shown in FIG. 2, it is assumed the total data duration for the time series of each asset's data is n, so that the sizes may be, for example, n, n/2, n/4, wherein each size represents a different time scale [generating a first windowed subsequence and a second windowed subsequence from multivariate time series data].”).
generating a first dependency graph object and a second dependency graph object from the first windowed subsequence and the second windowed subsequence…; (Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 constructs a segment similarity graph for each scale using the distance matrix [generating a first dependency graph object and a second dependency graph object from the first windowed subsequence and the second windowed subsequence…;].”).
determining a segmentation timestamp based on when a segment change occurred based on comparing the first dependency graph object and the second dependency graph object utilizing a similarity model…; (Kulkarni, ⁋31, “FIG. 3B is a segment graph 350 with segments from four different assets, namely, assets 1-4. The segment graph in FIG. 3B shows anomalous segment closeness between the assets. In particular, segments of asset 3 are closely grouped with a segment of asset 2, and a segment of asset 4 is closely grouped with a segment of asset 2; grouping the anomalous segments closer together based on closeness is interpreted as determining an anomalous timesteps or points in the dataset (i.e. determining a segmentation timestamp based on when a segment change occurred based on comparing the first dependency graph object and the second dependency graph object).”, and Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 [utilizing a similarity model…;] constructs a segment similarity graph for each scale using the distance matrix.”).
and generating a segmented multivariate time series by segmenting the multivariate time series data based on the segmentation timestamp. (Kulkarni, ⁋31, “FIG. 3B is a segment graph 350 with segments from four different assets, namely, assets 1-4. The segment graph in FIG. 3B shows anomalous segment closeness between the assets. In particular, segments of asset 3 are closely grouped with a segment of asset 2, and a segment of asset 4 is closely grouped with a segment of asset 2; grouping segments closer together based on anomalous closeness is interpreted as generating a segmented multivariate time series using segmentation timesteps, or anomalous points in the data (i.e. and generating a segmented multivariate time series by segmenting the multivariate time series data based on the segmentation timestamp.).”).
While Kulkarni teaches a system that segments multivariate time series data using graph objects, Kulkarni does not explicitly teach:
…utilizing a sparse graph recovery model…
…wherein the similarity model compares corresponding correlations of the first dependency graph object and the second dependency graph object to identify a change associated with a segment change to determine the segmentation timestamp…
Shrivastava teaches …utilizing a sparse graph recovery model… (Shrivastava, abstract, “We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data […utilizing a sparse graph recovery model…].”).
Kulkarni and Shrivastava are both in the same field of endeavor (i.e. graph analysis). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni and Shrivastava to teach the above limitation(s). The motivation for doing so is that recovering sparse conditional independence graphs improves understanding of relationships between different assets (cf. Shrivastava, pg. 1, “Recovering sparse conditional independence graphs from data is a fundamental problem in high dimensional statistics and time series analysis, and it has found applications in diverse areas. In computational biology, a sparse graph structure between gene expression data may be used to understand gene regulatory networks; in finance, a sparse graph structure between financial time series may be used to understand the relationship between different financial assets.”).
While the combination teaches a system for segmenting multivariate time series data, the combination does not explicitly teach:
…wherein the similarity model compares corresponding correlations of the first dependency graph object and the second dependency graph object to identify a change associated with a segment change to determine the segmentation timestamp…
Sulem teaches …wherein the similarity model compares corresponding correlations of the first dependency graph object and the second dependency graph object to identify a change associated with a segment change to determine the segmentation timestamp… (Sulem, pg. 2, “For multivariate time series, change-point detection is a task that has been widely studied in various settings (e.g., nonparametric [9], high-dimensional [10] or online [11]). The equivalent task for dynamic networks is often denoted network change-point detection (NCPD) and has recently become a popular problem with numerous successful ap plications in finance [7], neuroscience [12] or transport networks [13].”, and Sulem, pg. 5, “In our NCPD method [to identify a change associated with a segment change to determine the segmentation timestamp…], we propose a s-GNN architecture summarized in Figure 1, for learning a similarity score s(Gt1 ,Gt2 ) in [0,1] on the space of graph snapshots (G1,G2,...,Gt,...) from the dynamic network [compares corresponding correlations of the first dependency graph object and the second dependency graph object]. For this purpose, we design a similarity module […wherein the similarity model] for comparing the node-level embeddings output by a generic graph encoder (e.g., a graph convolutional network [40], a graph attention network [57], a GraphSage network [58] or a graph isomorphism network (GIN) [59]).”).
Kulkarni, in view of Shrivastava, and Sulem are both in the same field of endeavor (i.e. time series segmentation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava, and Sulem to teach the above limitation(s). The motivation for doing so is that using a siamese graph neural network as a similarity model improves the robustness of the similarity analysis between graph structures (cf. Sulem, pg. 2, “We propose a graph similarity learning model based on a siamese graph neural network able to handle any available node attributes, and demonstrate how it can be leveraged for the online NCPD problem with an adequate training procedure. In particular, our learnt similarity function is sensitive to both local and global displacements in the graph structure, and can effectively be employed in the context of change-point (and anomaly) detection in temporal networks.”).
Regarding claim 19, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 17. Kulkarni further teaches wherein the window size corresponds to a non-overlapping window, and wherein the first windowed subsequence and the second windowed subsequence include non-duplicate data. (Kulkarni, ⁋22, “The multi-scale segmentation module 104 creates segments of the time series datasets 102 of different sizes [wherein the window size corresponds to a non-overlapping window,]. FIG. 2 depicts a non-limiting example of multi-scale segmentation 200 in accordance with one or more example embodiments. As shown in FIG. 2, it is assumed the total data duration for the time series of each asset's data is n, so that the sizes may be, for example, n, n/2, n/4, wherein each size represents a different time scale [and wherein the first windowed subsequence and the second windowed subsequence include non-duplicate data.].”).
Claims 4 are rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, et al., US Pre-Grant Publication 2020/0372024A1 (“Kulkarni”) in view of Shrivastava, et al., Non-Patent Literature “GLAD: LEARNING SPARSE GRAPH RECOVERY” (“Shrivastava”) and further in view of Sulem, et al., Non-Patent Literature “Graph similarity learning for change-point detection in dynamic networks” (“Sulem”) and Runestone, Non-Patent Literature “An Adjacency Matrix” (“Runestone”).
Regarding claim 4, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1. While the combination teaches generating the dependency graph objects from the windowed subsequences as seen in claim 1, the combination does not explicitly teach includes generating an adjacency matrix indicating partial correlations between corresponding nodes and edges between two dependency graph objects.
Runestone teaches includes generating an adjacency matrix indicating partial correlations between corresponding nodes and edges between two dependency graph objects. (Runestone, pg. 1, “One of the easiest ways to implement a graph is to use a two-dimensional matrix. In this matrix implementation, each of the rows and columns represent a vertex in the graph. The value that is stored in the cell at the intersection of row v and column w indicates if there is an edge from vertex v to vertex w. When two vertices are connected by an edge, we say that they are adjacent [includes generating an adjacency matrix]. Figure 3 illustrates the adjacency matrix for the graph in Figure 2. A value in a cell represents the weight of the edge from vertex v to vertex w [indicating partial correlations between corresponding nodes and edges between two dependency graph objects.].”).
Kulkarni, in view of Shrivastava and Sulem, and Runestone are both in the same field of endeavor (i.e. graph analysis). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava and Sulem, and Runestone to teach the above limitation(s). The motivation for doing so is that using an adjacency matrix provides a simplified representation of graph connections (cf. Runestone, pg. 1, “One of the easiest ways to implement a graph is to use a two-dimensional matrix. In this matrix implementation, each of the rows and columns represent a vertex in the graph.”).
Claims 5, 12, 14-15, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, et al., US Pre-Grant Publication 2020/0372024A1 (“Kulkarni”) in view of Shrivastava, et al., Non-Patent Literature “GLAD: LEARNING SPARSE GRAPH RECOVERY” (“Shrivastava”) and further in view of Sulem, et al., Non-Patent Literature “Graph similarity learning for change-point detection in dynamic networks” (“Sulem”) and Brownlee, Non-Patent Literature “A Gentle Introduction to Mini-Batch Gradient Descent and How to Configure Batch Size” (“Brownlee”).
Regarding claim 5, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1. Kulkarni further teaches further comprising generating multiple dependency graph objects from the windowed subsequences… (Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 constructs a segment similarity graph for each scale using the distance matrix [further comprising generating multiple dependency graph objects from the windowed subsequences].”).
Shrivastava further teaches …that utilizes one instance of the sparse graph recovery model and shared parameters. (Shrivastava, abstract, “We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data […that utilizes one instance of the sparse graph recovery model and shared parameters.].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Shrivastava with the teachings of Kulkarni and Sulem for the same reasons disclosed in claim 1.
While the combination teaches a multivariate time series segmentation system using a sparse graph recovery model, the combination does not explicitly teach …at a same time as a batch operation…
Brownlee teaches …at a same time as a batch operation… (Brownlee, pg. 2, “Batch gradient descent is a variation of the gradient descent algorithm that calculates the error for each example in the training dataset, but only updates the model after all training examples have been evaluated. One cycle through the entire training dataset is called a training epoch. Therefore, it is often said that batch gradient descent performs model updates at the end of each training epoch […at a same time as a batch operation…].”).
Kulkarni, in view of Shrivastava and Sulem, and Brownlee are both in the same field of endeavor (i.e. machine learning). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava and Sulem, and Brownlee to teach the above limitation(s). The motivation for doing so is that batch training improves the stability of the model convergence (cf. Brownlee, pg. 3, “The decreased update frequency results in a more stable error gradient and may result in a more stable convergence”).
Regarding claim 12, the claim is similar to claim 5 and rejected under the same rationales.
Regarding claim 14, Kulkarni in view of Shrivastava, Sulem, and Brownlee teaches the system of claim 12. Shrivastava further teaches wherein: the sparse graph recovery model generates dependency graph objects that exhibit partial correlation between variables; and the dependency graph objects include conditional independence graph objects. (Shrivastava, pg. 1, “Recovering sparse conditional independence graphs from data is a fundamental problem in high dimensional statistics and time series analysis [wherein: the sparse graph recovery model generates dependency graph objects]”, and Shrivastava, pg. 2, “Given m observations of ad-dimensional multivariate Gaussian random variable X = [X1,...,Xd]⊤, the sparse graph recovery problem aims to estimate its covariance matrix Σ∗ and precision matrix Θ∗ =(Σ∗)−1. The ij-th component of Θ∗ is zero if and only if Xi and Xj are conditionally independent given the other variables {Xk}k̸=i,j [that exhibit partial correlation between variables; and the dependency graph objects include conditional independence graph objects.].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Shrivastava with the teachings of Kulkarni, Sulem, and Brownlee for the same reasons disclosed in claim 12.
Regarding claim 15, Kulkarni in view of Shrivastava, Sulem, and Brownlee teaches the system of claim 12. Kulkarni further teaches wherein generating the dependency graph objects from the windowed subsequences includes generating a visual graph of nodes and edges, where the edges indicate a positive or a negative partial correlation between connected nodes. (Kulkarni, ⁋25, “FIG. 3A shows an example of a segment graph 300 [wherein generating the dependency graph objects from the windowed subsequences includes generating a visual graph of nodes and edges,] wherein the vertices represent data coming from all assets for a given-time scale, and the edges represent the distance between the distributions of the data represented by the vertices [where the edges indicate a positive or a negative partial correlation between connected nodes.].”).
Regarding claim 20, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 17. Kulkarni further teaches further comprising generating the first dependency graph object and the second dependency graph object…based on the first windowed subsequence and the second windowed subsequence. (Kulkarni, ⁋25, “The multi-scale anomalous asset detection module 106 constructs a segment similarity graph for each scale using the distance matrix [further comprising generating the first dependency graph object and the second dependency graph object…based on the first windowed subsequence and the second windowed subsequence.].”).
Shrivastava further teaches …utilizing the sparse graph recovery model… (Shrivastava, abstract, “We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data […utilizing the sparse graph recovery model…].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Shrivastava with the teachings of Kulkarni and Sulem for the same reasons disclosed in claim 17.
While the combination teaches a multivariate time series segmentation system using a sparse graph recovery model, the combination does not explicitly teach …in a single batch operation…
Brownlee teaches …in a single batch operation… (Brownlee, pg. 2, “Batch gradient descent is a variation of the gradient descent algorithm that calculates the error for each example in the training dataset, but only updates the model after all training examples have been evaluated. One cycle through the entire training dataset is called a training epoch. Therefore, it is often said that batch gradient descent performs model updates at the end of each training epoch […in a single batch operation…].”).
Kulkarni, in view of Shrivastava and Sulem, and Brownlee are both in the same field of endeavor (i.e. machine learning). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava and Sulem, and Brownlee to teach the above limitation(s). The motivation for doing so is that batch training improves the stability of the model convergence (cf. Brownlee, pg. 3, “The decreased update frequency results in a more stable error gradient and may result in a more stable convergence”).
Claims 7, 16, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, et al., US Pre-Grant Publication 2020/0372024A1 (“Kulkarni”) in view of Shrivastava, et al., Non-Patent Literature “GLAD: LEARNING SPARSE GRAPH RECOVERY” (“Shrivastava”) and further in view of Sulem, et al., Non-Patent Literature “Graph similarity learning for change-point detection in dynamic networks” (“Sulem”) and Tan, et al., US Pre-Grant Publication 2024/0160160A1 (“Tan”).
Regarding claim 7, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 1.
While the combination teaches a multivariate time series segmentation system using a sparse graph recovery model, the combination does not explicitly teach further comprising: generating, using a refined window size, additional windowed subsequences from the multivariate time series data based on the one or more segmentation timestamps, wherein the refined window size is smaller than the window size; determining one or more refined segmentation timestamps from the additional windowed subsequences; and refining locations of segments within the segmented multivariate time series based on the one or more refined segmentation timestamps.
Tan teaches:
further comprising: generating, using a refined window size, additional windowed subsequences from the multivariate time series data based on the one or more segmentation timestamps, wherein the refined window size is smaller than the window size; (Tan, ⁋32, “A sliding time-window is defined which slides inside an internal greater time window [wherein the refined window size is smaller than the window size;]. The sliding time window is applied to all algorithms, and all CPs over all algorithms are summed up. The same procedure is performed with further sliding time windows [further comprising: generating, using a refined window size, additional windowed subsequences from the multivariate time series data based on the one or more segmentation timestamps,]. Those sliding time windows, i.e., the CPs of these windows, are presented to the user that contain a high number of CPs.”).
determining one or more refined segmentation timestamps from the additional windowed subsequences; (Tan, ⁋56, “According to an embodiment, the selection out of the candidate CPs is according one of the following: (i) selecting the CPs randomly out of the candidate CPs, (ii) defining a sliding time-window length smaller than the time-window, summing up the number of CP candidates detected across all CP algorithms, and selecting windows with a high sum are selected [determining one or more refined segmentation timestamps from the additional windowed subsequences;]”).
and refining locations of segments within the segmented multivariate time series based on the one or more refined segmentation timestamps. (Tan, ⁋16, “CPD can be used to cut the otherwise continuous signal into segments with start (a first change point) and end (the next change point) times. These segments can be used as samples in an anomaly detection or in a classification process [and refining locations of segments within the segmented multivariate time series based on the one or more refined segmentation timestamps.]”).
Kulkarni, in view of Shrivastava and Sulem, and Tan are both in the same field of endeavor (i.e. data segmentation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava and Sulem, and Tan to teach the above limitation(s). The motivation for doing so is that finding additional change points increases the understanding of time series data (cf. Tan, ⁋17, “Similar to process monitoring by anomaly detection or classification, the CPD is used to split a continuous signal into several meaningful segments.”).
Regarding claim 16, the claim is similar to claim 7 and rejected under the same rationales.
Regarding claim 18, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 17.
While the combination teaches a multivariate time series segmentation system using a sparse graph recovery model, the combination does not explicitly teach wherein the window size corresponds to an overlapping window, and wherein the first windowed subsequence and the second windowed subsequence include duplicative data.
Tan teaches wherein the window size corresponds to an overlapping window, and wherein the first windowed subsequence and the second windowed subsequence include duplicative data. (Tan, ⁋32, “A sliding time-window is defined which slides inside an internal greater time window [wherein the window size corresponds to an overlapping window,]. The sliding time window is applied to all algorithms, and all CPs over all algorithms are summed up. The same procedure is performed with further sliding time windows [and wherein the first windowed subsequence and the second windowed subsequence include duplicative data.]. Those sliding time windows, i.e., the CPs of these windows, are presented to the user that contain a high number of CPs.”).
Kulkarni, in view of Shrivastava and Sulem, and Tan are both in the same field of endeavor (i.e. data segmentation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava and Sulem, and Tan to teach the above limitation(s). The motivation for doing so is that finding additional change points increases the understanding of time series data (cf. Tan, ⁋17, “Similar to process monitoring by anomaly detection or classification, the CPD is used to split a continuous signal into several meaningful segments.”).
Claims 10 are rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, et al., US Pre-Grant Publication 2020/0372024A1 (“Kulkarni”) in view of Shrivastava, et al., Non-Patent Literature “GLAD: LEARNING SPARSE GRAPH RECOVERY” (“Shrivastava”) and further in view of Sulem, et al., Non-Patent Literature “Graph similarity learning for change-point detection in dynamic networks” (“Sulem”) and Ottersten, et al., Non-Patent Literature “Accurate Changing Point Detection for ℓ1 Mean Filtering” (“Ottersten”).
Regarding claim 10, Kulkarni in view of Shrivastava and Sulem teaches the computer-implemented method of claim 8.
While the combination teaches the second order distance and a similarity model, the combination does not explicitly teach wherein the allocation algorithm further comprises: reducing the…distance by a filtering out sequence values below a noise threshold to generate a filtered sequence; and traversing the filtered sequence for non-zero values to identify the one or more segmentation timestamps.
Ottersten teaches wherein the allocation algorithm further comprises: reducing the…distance by a filtering out sequence values below a noise threshold to generate a filtered sequence; and traversing the filtered sequence for non-zero values to identify the one or more segmentation timestamps. (Ottersten, pg. 298 col. 1, “This is referred to as ℓ1 mean filtering [18] which is also known as the total variation (TV) denoising [5] problem. The penalty in equation (1) forces xi to be constant and we obtain a piece-wise constant solution; filtering out noise is interpreted as reducing a distance as removing noisy values improves the effectiveness of data comparisons (i.e. reducing the…distance by a filtering out sequence values below a noise threshold to generate a filtered sequence;). The ℓ1 mean filtering problem is used to detect changes in the mean of the time series yi. The non-zero entries in the sparse changing vector indicate the changing points in the time series yi [and traversing the filtered sequence for non-zero values to identify the one or more segmentation timestamps.].”).
Kulkarni, in view of Shrivastava and Sulem, and Ottersten are both in the same field of endeavor (i.e. data segmentation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava and Sulem, and Ottersten to teach the above limitation(s). The motivation for doing so is that removing noise from data improves the ability to find trends in the data (cf. Ottersten, pg. 297 col. 1, “Finding the underlying linear trends in time series data is a common signal processing problem that arises in many applications in areas such as financial time series analysis [1], in biological and medical sciences [2] and climatology [3] to list a few. This is a difficult estimation problem when noise and imperfections are present in the signal measurements. The main challenge in mean and trend filtering is to find the so-called changing points in the trend.”).
Claims 13 are rejected under 35 U.S.C. 103 as being unpatentable over Kulkarni, et al., US Pre-Grant Publication 2020/0372024A1 (“Kulkarni”) in view of Shrivastava, et al., Non-Patent Literature “GLAD: LEARNING SPARSE GRAPH RECOVERY” (“Shrivastava”) and further in view of Sulem, et al., Non-Patent Literature “Graph similarity learning for change-point detection in dynamic networks” (“Sulem”), Brownlee, Non-Patent Literature “A Gentle Introduction to Mini-Batch Gradient Descent and How to Configure Batch Size” (“Brownlee”), and Chen, et al., Non-Patent Literature “Unsupervised Multimodal Change Detection Based on Structural Relationship Graph Representation Learning” (“Chen”).
Regarding claim 13, Kulkarni in view of Shrivastava, Sulem, and Brownlee teaches the system of claim 12.
While the combination teaches a sparse graph recovery model using batch learning, the combination does not explicitly teach wherein the sparse graph recovery model is an unsupervised deep-learning sparse graph recovery model trained to generate batches of dependency object graphs.
Chen teaches wherein the sparse graph recovery model is an unsupervised deep-learning sparse graph recovery model trained to generate batches of dependency object graphs. (Chen, abstract, “Unsupervised multimodal change detection is a practical and challenging topic that can play an important role in time-sensitive emergency applications…First, structural graphs are generated from preprocessed multimodal image pairs by means of an object-based image analysis approach. Then, a structural relationship graph convolutional autoencoder (SR-GCAE) is proposed to learn robust and representative features from graphs. Two loss functions aiming at reconstructing vertex information and edge information are presented to make the learned representations applicable for structural relationship similarity measurement [wherein the sparse graph recovery model is an unsupervised deep-learning sparse graph recovery model trained to generate batches of dependency object graphs.].”).
Kulkarni, in view of Shrivastava, Sulem, and Brownlee, and Chen are both in the same field of endeavor (i.e. change detection). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Kulkarni, in view of Shrivastava, Sulem, and Brownlee, and Chen to teach the above limitation(s). The motivation for doing so is that unsupervised sparse graph recovery model learns robust dependency graphs for analysis (cf. Chen, pg. 2 col. 1, “the proposed network can learn robust high-level graph representations to measure the similarity levels of local and nonlocal structural relationships.”).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Hallac, et al., “Network inference via the time-varying graphical lasso” discloses a system that uses a time-varying graphical lasso to perform segmentation of raw multivariate time series data.
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 NICHOLAS S WU whose telephone number is (571)270-0939. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Michelle Bechtold can be reached at 571-431-0762. 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.
/N.S.W./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights