DETAILED ACTION
1. This office action is in response to the Application No. 18065088 filed on 02/19/2026. Claims 22, 32 and 40 has been cancelled. Claims 21, 23-31, 33-41 and 42 are presented for examination and are currently pending. Applicant’s arguments have been carefully and respectfully considered.
Response to Arguments
2. Applicant’s arguments are moot in view of the new grounds of rejection. The Examiner is withdrawing the rejections in the previous office action 11/26/2025 because the applicant amendments necessitated the new grounds of rejection presented in this office action.
A new secondary reference has been applied to address the newly added limitations.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
3. Claims 21, 23-31, 33-41 and 42 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claims 21, 33 and 36 recites “encodes a stored, node-pair-specific Euclidean distance representing a measured geometric separation between the two nodes”.
The limitations “encodes a stored, node-pair-specific Euclidean distance representing a measured geometric separation between the two nodes” does not have basis in the original disclosure. The Applicant’s specification at [0084] stated that “The exemplary DSP structure may further include a discriminant layer which applies discriminant function analysis to find a linear combination of features that characterizes or separates two or more groups of data”. This citation in the specification is different from the claim limitation “encodes a stored, node-pair-specific Euclidean distance representing a measured geometric separation between the two nodes”.
Furthermore, there is no single occurrence of the limitation “encode” in the specification.
Therefore, the claim limitation “encodes a stored, node-pair-specific Euclidean distance representing a measured geometric separation between the two nodes” does not have support in the instant specification.
Claims 23-31, 34, 35, 37-39, 41 and 42 that are not specifically mentioned are rejected due to dependency.
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.
4. Claims 21, 23-25, 27, 28, 33, 35, 36 and 42 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. ("Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting." arXiv:1709.04875v4 [cs.LG] 12 Jul 2018) in view of Shuman et al. ("The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains." IEEE signal processing magazine 30.3 (2013): 83-98, arXiv:1211.0053v2 [cs.DM] 10 Mar 2013) in view of Evans et al (US20040158569) and further in view of Cao et al. ("Generalized Multi-View Embedding for Visual Recognition and Cross-Modal Retrieval," in IEEE Transactions on Cybernetics, vol. 48, no. 9, pp. 2542-2555, Date of publication September 6, 2017)
Regarding claim 21, Yu teaches a system comprising: one or more processors; and at least one memory storing processor-executable instructions that, when executed by any one or more of the one or more processors, causes the one or more processors to perform operations comprising (All experiments are compiled and tested on a Linux cluster (CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz, GPU: NVIDIA GeForce GTX 1080), pg. 5, left col., section 4.3):
receive irregular grid graph data (In this work, we define the traffic network on a graph and focus on structured traffic time series (pg. 2, left col. last para.); Accordingly, in our model, the graph convolution is employed directly on graph-structured data to extract highly meaningful patterns and features in the space domain, pg. 3, left col., Section 3.2; Figure 1: Graph-structured traffic data. Each vt indicates a frame of current traffic status at time step t, which is recorded in a graph-structured data matrix. The Examiner noes graph traffic time series data is an irregular grid graph data);
generating, using an optimized digital signal processor (DSP) (Therefore, the data point vt can be regarded as a graph signal … as shown in Figure 1 (pg. 2, right col., first para.); All experiments are compiled and tested on a Linux cluster (CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz, GPU: NVIDIA GeForce GTX 1080), pg. 5, left col. section 4.3),
a prediction … based at least in part on the irregular grid graph data (Figure 2: Architecture of spatio-temporal graph convolutional networks … The input vt−M+1, ..., vt is uniformly processed by ST-Conv blocks to explore spatial and temporal dependencies coherently. Comprehensive features are integrated by an output layer to generate the final prediction v, pg. 3, left col., Fig. 2),
wherein: the optimized DSP comprises a set of parameters for: (i) a set of hidden layers (Figure 2: Architecture of spatio-temporal graph convolutional networks. The framework STGCN consists of two spatio-temporal convolutional blocks (ST-Conv blocks) and a fully-connected output layer in the end. Each ST-Conv block contains two temporal gated convolution layers and one spatial graph convolution layer in the middle, pg. 3, left col. Fig. 2),
configured to generate an output (Comprehensive features are integrated by an output layer to generate the final prediction vˆ, pg. 3, left col., Fig. 2),
wherein a hidden layer of the set of hidden layers comprises a set of kernels (HKs) (To localize the filter and reduce the number of parameters, the kernel Θ can be restricted to a polynomial of Λ as Θ(Λ) = PK−1 k=0 θkΛ k , where θ ∈ R K is a vector of polynomial coefficients. K is the kernel size of graph convolution, which determines the maximum radius of the convolution from central nodes, pg. 3, left col. last para.), and
associated with one or more initial Laplacian operators and corresponding initial filter parameters (A layer-wise linear formulation can be defined by stacking multiple localized graph convolutional layers with the first-order approximation of graph Laplacian (pg. 3, right col. first para.); where graph Fourier basis U ∈ R n×n is the matrix of eigenvectors of the normalized graph Laplacian L = In − D− 1 2 W D− 1 2 = UΛU T ∈ R n×n (In is an identity matrix, D ∈ R n×n is the diagonal degree matrix with Dii = ΣjWij ); Λ ∈ R n×n is the diagonal matrix of eigenvalues of L, and filter Θ(Λ) is also a diagonal matrix. By this definition, a graph signal x is filtered by a kernel Θ with multiplication between Θ and graph Fourier transform UT x, pg. 2, section 2.2); and
Yu does not explicitly teach receiving irregular grid graph data that comprise a set of nodes and an edge that (i) connects two nodes of the set of nodes and (ii) encodes a stored, node-pair-specific Euclidean distance representing a measured geometric separation between the two nodes; and generating, a prediction for a node of the set of nodes based at least in part on the irregular grid graph data, heterogeneous kernels (HKs), an HK of the set of HKs comprises a corresponding set of filters selected from a constructed set of filters, and(ii) a discriminant layer configured to apply a discriminant function analysis to the output of the set of hidden layers to find a linear combination of features that separates two or more groups of data within the irregular grid graph data to generate the prediction for the node of the set of nodes.
Shuman teaches receiving irregular grid graph data (highlight the importance of incorporating the irregular structures of graph data domains when processing signals on graphs, abstract) that comprise a set of nodes and an edge that (i) connects two nodes of the set of nodes (We are interested in analyzing signals defined on an undirected, connected, weighted graph G = {V,E,W}, which
consists of a finite set of vertices V with |V| = N, a set of edges E, and a weighted adjacency matrix W. If there is an edge e =(i,j) connecting vertices i and j, the entry Wi,j represents the weight of the edge, pg. 3, section A. Weighted Graphs and Graph Signals) and
(ii) encodes a stored (The graph spectral domain designs of graph wavelets are based on the spectral features of the graph, which are encoded, e.g., in the eigenvalues and eigenvectors of one of the graph matrices, pg. 11, section B. Graph Spectral Domain Designs), node-pair-specific Euclidean distance representing a measured geometric separation between the two nodes (one common way to define the weight of an edge connecting vertices i and j is via a thresholded Gaussian kernel weighting function: ... for some parameters θ and κ. In (1), dist(i,j) may represent a physical distance between vertices i and j, or the Euclidean distance between two feature vectors describing i and j, pg. 3, section A. Weighted Graphs and Graph Signals); and
generating, a prediction for a node of the set of nodes based at least in part on the irregular grid graph data (the vertex set is first partitioned into sets of even and odd nodes, V = VO∪VE. Each odd node computes its prediction coefficient using its own data and data from its even neighbors, pg. 10 right col., last para to page 11, left col., first para.),
a linear combination of features that separates two or more groups of data within the irregular grid graph data (2) Filtering in the Vertex Domain: To filter a signal in the vertex domain, we simply write the output fout(i) at vertex I a linear combination of the components of the input signal at vertices within a K-hop local neighborhood of vertex i:, pg. 8, right col., last para.) to generate the prediction for the node of the set of nodes (Then each even node computes its update coefficients using its own data and the prediction coefficients of its neighboring odd nodes, pg. 11, left col., first para.)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to have modified the apparatus of Yu to incorporate the teachings of Shuman in order to efficiently extract information from high-dimensional data on graphs (Shuman, abstract).
Modified Yu does not explicitly teach heterogeneous kernels (HKs), an HK of the set of HKs comprises a corresponding set of filters selected from a constructed set of filters, (ii) a discriminant layer configured to apply a discriminant function analysis to the output of the set of hidden layers to find a linear combination of features that separates two or more groups of data within the irregular grid graph data to generate the prediction ...
Evans teaches heterogeneous kernels (HKs), an HK of the set of HKs comprises a corresponding set of filters selected from a constructed set of filters (the component filters can be composed of heterogeneous combinations of filter types [0115]).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to have modified the apparatus of Modified Yu to incorporate the teachings of Evans for the benefit of an alternative modelling strategy using stacked filters in a neural network as a means for creating an aggregation function (Evans [0133])
Modified Yu does not explicitly teach (ii) a discriminant layer configured to apply a discriminant function analysis to the output of the set of hidden layers to find a linear combination of features that separates two or more groups of data within the irregular grid graph data to generate the prediction.
Cao teaches (ii) a discriminant layer (For the nonlinear mapping by neural networks, we deploy a linear embedding layer on top of the networks. This scheme is illustrated in Fig. 2, pg. 2546, left col., second para.)
configured to apply a discriminant function analysis (H. Multi-View Discriminant Analysis: MvDA is the multi-view version of LDA which maximizes the ratio of the determinant of the between-class scatter matrix to that of the within-class scatter matrix. Its objective function is …, pg. 4, left col., second to the last para.) to the output of the set of hidden layers (we took three hidden layer, each with 1024 neurons with the relu activation before the 50-D linear embedding layer (pg. 2549, right col., last para. The Examiner notes that the linear embedding layer is applied to the output of the three hidden layers); we train multiple neural networks whose outputs are connected to the linear layer, pg. 2546, left col., second para.)
to find a linear combination of features that separates two or more groups of data (Taking the class discrimination into consideration, the novel MvMDA extends to separate the data of different classes between views while making the intraclass data compact., pg. 2546, left col., second to the last para.) within the irregular grid graph data to generate the prediction (The prediction of target classes is undertaken on a hypergraph to better integrate different views, pg. 2549, left col., last para.).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to have modified the apparatus of Modified Yu to incorporate the teachings of Cao for the benefit of multi-view embedding method using the graph embedding framework by exploiting the distances between class centers of different views (Cao, conclusion)
Regarding claim 23, Modified Yu teaches the system of claim 21, Yu teaches wherein the set of nodes comprise: (i) a known graph dataset (We verify our model on two real-world traffic datasets, BJER4 and PeMSD7, collected by Beijing Municipal Traffic Commission and California Deportment of Transportation, respectively, pg. 4, right col., section 4) comprising
one or more known graph nodes corresponding to one or more known target data (Specifically, the input of traffic prediction is composed of M frame of road graphs as Figure 1 shows. Each frame vt can be regarded as a matrix whose column i is the Ci-dimensional value of vt at the i th node in graph Gt, as X ∈ R n×Ci (in this case, Ci = 1)(pg. 3, right col., last para.); Figure 4: Speed prediction in the morning peak and evening rush hours of the dataset PeMSD7, pg. 6, left col.), and
(ii) an unknown graph dataset comprising one or more unknown graph nodes without one or more corresponding target data (The time period used is from 1st July to 31st August, 2014 except the weekends. We select the first month of historical speed records as training set, and the rest serves as validation and test set respectively, pg. 4, right col., section 4.1. The Examiner notes the test set as the unknown dataset with no corresponding prediction).
Regarding claim 24, Modified Yu teaches the system of claim 23, Yu teaches wherein the optimized DSP is previously trained using the known graph dataset (We train our models by minimizing the mean square error using RMSprop for 50 epochs with batch size as 50, pg. 5, right col., second para.); The time range of PeMSD7 dataset is in the weekdays of May and June of 2012, pg. 5, left col., first para.).
Regarding claim 25, Modified Yu teaches the system of claim 23, Yu teaches wherein the prediction (we show their predictions during morning peak and evening rush hours, as shown in Figure 4. It is easy to observe that our proposal STGCN captures the trend of rush hours more accurately than other methods, pg. 6, left col., first para.) comprises
a machine learning based prediction for an unknown graph node of the unknown graph dataset (The time period used is from 1st July to 31st August, 2014 except the weekends. We select the first month of historical speed records as training set, and the rest serves as validation and test set respectively, pg. 4, right col., section 4.1. The Examiner notes the test set as the unknown dataset).
Regarding claim 27, Modified Yu teaches the system of claim 21, Evans teaches wherein the HK is previously generated based at least in part on a weighted combination of the corresponding set of filters (One such measure of filter complexity (though not limited to) could relate to the size of the filter in terms of bytes used to store the filter. These measures could be combined using a weighted sum, whereby the weights are selected by the user. The approach with the highest combined score would be selected [0212]).
The same motivation to combine as the independent claim 21 applies here.
Regarding claim 28, Modified Yu teaches the system of claim 21, Evans teaches wherein the hidden layer of the set of hidden layers is previously generated based at least in part on a weighted combination of the set of HKs (One such measure of filter complexity (though not limited to) could relate to the size of the filter in terms of bytes used to store the filter. These measures could be combined using a weighted sum [0212]).
The same motivation to combine as the independent claim 21 applies here.
Regarding claim 33, claim 33 is similar to claim 21. It is rejected in the same manner and reasoning applying.
Regarding claim 35, claim 35 is similar to claim 27. It is rejected in the same manner and reasoning applying.
Regarding claim 36, claim 36 is similar to claim 21. It is rejected in the same manner and reasoning applying.
Regarding claim 42, Modified Yu teaches the system of claim 21, Yu teaches wherein the set of parameters comprises the initial filter parameters (To localize the filter and reduce the number of parameters, the kernel Θ can be restricted to a polynomial, pg. 3, left col., last para.),
an initial filter number (a graph signal x is filtered by a kernel Θ (pg. 2, right
col., second to the last para.); In this scenario, K is the number of successive
filtering operations or convolutional layers in a model instead, pg. 3, right col., second to the last para.), an initial HK parameter (where θ0, θ1 are two shared parameters of the kernel, pg. 3, right col., first para.), an initial HK number (K is the kernel size of graph convolution, pg. 3, left col., last para.), and
an initial hidden layer number (As illustrated in Figure 2 (mid), the spatial layer in the middle is to bridge two temporal layers which can achieve fast spatial-state propagation from graph convolution through temporal convolutions, pg. 4, left col., second to the last para.).
5. Claims 26 and 34 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. ("Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting." arXiv:1709.04875v4 [cs.LG] 12 Jul 2018) in view of Shuman et al. ("The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains." IEEE signal processing magazine 30.3 (2013): 83-98, arXiv:1211.0053v2 [cs.DM] 10 Mar 2013) in view of Evans et al (US20040158569) in view of Cao et al. ("Generalized Multi-View Embedding for Visual Recognition and Cross-Modal Retrieval," in IEEE Transactions on Cybernetics, vol. 48, no. 9, pp. 2542-2555, Date of publication September 6, 2017) and further in view of Du et al. (“Topology adaptive graph convolutional networks.”, arXiv:1710.10370v5 [cs.LG] 11 Feb 2018)
Regarding claim 26, Modified Yu teaches the system of claim 21, Yu teaches wherein the corresponding set of filters comprise: (i) at least one of: a K-order Chebyshev filter type (The graph convolution can then be rewritten as,
PNG
media_image1.png
138
502
media_image1.png
Greyscale
pg. 3, left col., left col., last para.),
(ii) a first order renormalized filter type (1 st-order Approximation- A layer-wise linear formulation can be defined by stacking multiple localized graph convolutional layers with the first-order approximation of graph Laplacian … Due to the scaling and normalization in neural networks, we can further assume that λmax ≈ 2. Thus, the Eq. (3) can be simplified to, … where θ0, θ1 are two shared parameters of the kernel. In order to constrain parameters and stabilize numerical performances, θ0 and θ1 are replaced by a single parameter θ by letting θ = θ0 = −θ1; W and D are renormalized by W˜ = W + In and D˜ ii = ΣjW˜ ij separately, pg. 3, right col., second to the last para.), and
Modified Yu does not explicitly teach (iii) at least one third filter associated with a K-order topology adaptive filter type.
Du teaches (iii) a K-order topology adaptive filter type (The proposed TAGCN explores a general K-localized filter for graph convolution in the vertex domain to extract local features on a set of size-1 up to size-K receptive fields. The topologies of these filters are adaptive to the topology of the graph as they scan the graph to perform convolution, pg. 2, first bullet point).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Modified Yu to incorporate the method of Du for the benefit of designing a set of fixed-size learnable filters to perform convolutions on graphs (Du, abstract)
Regarding claim 34, claim 34 is similar to claim 26. It is rejected in the same manner and reasoning applying.
6. Claims 29, 30, 37 and 38 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. ("Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting." arXiv:1709.04875v4 [cs.LG] 12 Jul 2018) in view of Shuman et al. ("The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains." IEEE signal processing magazine 30.3 (2013): 83-98, arXiv:1211.0053v2 [cs.DM] 10 Mar 2013) in view of Evans et al (US20040158569) in view of Cao et al. ("Generalized Multi-View Embedding for Visual Recognition and Cross-Modal Retrieval," in IEEE Transactions on Cybernetics, vol. 48, no. 9, pp. 2542-2555, Date of publication September 6, 2017) and further in view of Li et al. ("Learning graph while training: An evolving graph convolutional neural network." arXiv preprint arXiv:1708.04675 (2017))
Regarding claim 29, Modified Yu teaches the system of claim 21, Modified Yu does not explicitly teach wherein the optimized DSP is previously generated by updating the one or more initial Laplacian operators and the corresponding initial filter parameters in an iterative manner to optimize an objective function.
Li teaches wherein the optimized DSP is previously generated by updating the one or more initial Laplacian operators and the corresponding initial filter parameters in an iterative manner to optimize an objective function (In Fig. 2, the network consists of two SGC-LL layers, in which the two sets of graph Laplacian Lk will be updated independently and will probably diverge because the input xk is different, and the two layers worked on different feature maps (pg. 5, last para.); As shown in Fig. 4, the weighted l2 loss dropped quickly during the epoch 5∼20, so did the mean RMSE score. In the meanwhile, the graph Laplacians keep evolving according to gradient back-propagated from next layer (pg. 6, last para.); In proposed EGCN, the symmetric positive semi-definite matrix M = WdWTd is the trainable weight of SGC-LL layer. The Wkd ∈ R fk×fk works as a transform basis to some domain in which we measure the Euclidean distance between x and y, pg. 5, first para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Modified Yu to incorporate the method of Li for the benefit of acceleration of parameter fitting and the significantly improved prediction accuracy on multiple graph-structured datasets (Li, abstract).
Regarding claim 30, Modified Yu teaches the system of claim 29, Yu teaches wherein the objective function is a loss function (We use L2 loss to measure the performance of our model. Thus, the loss function of STGCN for traffic prediction can be written as … pg. 4, right col.).
Regarding claim 37, claim 37 is similar to claim 29. It is rejected in the same manner and reasoning applying.
Regarding claim 38, claim 38 is similar to claim 30. It is rejected in the same manner and reasoning applying.
7. Claims 31 and 39 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. ("Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting." arXiv:1709.04875v4 [cs.LG] 12 Jul 2018) in view of Shuman et al. ("The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains." IEEE signal processing magazine 30.3 (2013): 83-98, arXiv:1211.0053v2 [cs.DM] 10 Mar 2013) in view of Evans et al (US20040158569) in view of Cao et al. ("Generalized Multi-View Embedding for Visual Recognition and Cross-Modal Retrieval," in IEEE Transactions on Cybernetics, vol. 48, no. 9, pp. 2542-2555, Date of publication September 6, 2017) in view of Li et al. ("Learning graph while training: An evolving graph convolutional neural network." arXiv preprint arXiv:1708.04675 (2017)) and further in view of Bergstra (US20170140259)
Regarding claim 31, Modified Yu teaches the system of claim 29, Modified Yu does not explicitly teach wherein the objective function is a reward function.
Bergstra teaches wherein the objective function is a reward function (The two or more artificial neural networks can have different objective functions … an example objective function is a suitable reward function [0164]).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Modified Yu to incorporate the method of Bergstra for the benefit of optimizing the objective function with respect to the input (Bergstra [0062])
Regarding claim 39, claim 39 is similar to claim 31. It is rejected in the same manner and reasoning applying.
8. Claim 41 is rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. ("Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting." arXiv:1709.04875v4 [cs.LG] 12 Jul 2018) in view of Shuman et al. ("The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains." IEEE signal processing magazine 30.3 (2013): 83-98, arXiv:1211.0053v2 [cs.DM] 10 Mar 2013) in view of Evans et al. (US20040158569) in view of Cao et al. ("Generalized Multi-View Embedding for Visual Recognition and Cross-Modal Retrieval," in IEEE Transactions on Cybernetics, vol. 48, no. 9, pp. 2542-2555, Date of publication September 6, 2017) and further in view of Auddy et al. ("Discriminant vector tranformations in neural network classifiers." 2016 International Joint Conference on Neural Networks (IJCNN). IEEE, 2016).
Regarding claim 41, Modified Yu teaches the system of claim 21, Yu teaches irregular grid graph data (In this work, we define the traffic network on a graph and focus on structured traffic time series (pg. 2, left col. last para.); Accordingly, in our model, the graph convolution is employed directly on graph-structured data to extract highly meaningful patterns and features in the space domain, pg. 3, left col., Section 3.2; Figure 1: Graph-structured traffic data. Each vt indicates a frame of current traffic status at time step t, which is recorded in a graph-structured data matrix. The Examiner noes graph traffic time series data is an irregular grid graph data)
the discriminant function analysis comprises applying a sigmoid function to the output of the set of hidden layers (The sigmoid gate σ(Q) controls which input P of the current states are relevant for discovering compositional structure and dynamic variances in time series. The non-linearity gates contribute to the exploiting of the full input filed through stacked temporal layers as well, pg. 4, left col., third para.), and
Cao teaches (ii) responsive to the irregular grid graph data requiring non-linear or binary classification (We propose a unified solution for subspace learning methods using the Rayleigh quotient, which is extensible for multiple views, supervised learning, and nonlinear embeddings, abstract),
(iii) responsive to the irregular grid graph data requiring a regression operation (In order to correlate the labels with the image contents, we train a ridge regressor with 10-fold cross-validation to map the VGG-16 image features to each dimension of the word vectors, respectively. The regressor outputs are used as the class label features, pg. 2549, left col., second para.),
the discriminant function analysis comprises applying a linear activation function to the output of the set of hidden layers (On the COCO dataset, we chose a single hidden layer with 1500 relu neurons, and the dimensionality of the final linear layer is also 1500, pg. 2549, right col., last para.).
The motivation to combine independent claim 21 applies here.
Modified Yu does not explicitly teach wherein:(i) responsive to the irregular grid graph data requiring multiclass classification, the discriminant function analysis comprises applying a softmax function to the output of the set of hidden layers,
(iii) responsive to the irregular grid graph data requiring a regression operation, the discriminant function analysis comprises applying a linear activation function to the output of the set of hidden layers.
Auddy teaches wherein:(i) responsive to the data requiring multiclass classification (The output value for each class is called a discriminant and the class with the highest discriminant value is considered to be the correct class, pg. 1781, left col., first para.),
the discriminant function analysis comprises applying a softmax function to the output of the set of hidden layers (From Fig. 4, it is evident that the maximum value of the discriminant varies over a wide range and the probability of correct classification for a majority of these maximum discriminant values is very high. A softmax operation is used to scale these values so that they again sum to 1, pg.1784, right col., first para.),
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Modified Yu to incorporate the teachings of Auddy for the benefit of developing a method for mapping MLP discriminants into more accurate estimates of posterior probabilities, for cases having a large number of classes (Auddy, conclusion).
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 extension fee 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 date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MORIAM MOSUNMOLA GODO whose telephone number is (571)272-8670. The examiner can normally be reached Monday-Friday 8am-5pm 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 T Bechtold can be reached on (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.
/M.G./Examiner, Art Unit 2148
/MICHELLE T BECHTOLD/ Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights