DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Responsive to the communication dated 03/11/2026
Claims 1-22 are presented for examination
Information Disclosure Statement
The IDS dated 03/21/2026 and 04/27/2026 has been reviewed. See attached.
Finality
THIS ACTION IS MADE FINAL. 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 mailing date of this final action.
Response to Arguments -101
Applicant’s arguments, see pages 9-15, filed 03/11/2026, with respect to the rejection of claims 1-20 under 35 U.S.C. 101 have been fully considered and are persuasive. The rejection of claims 1-20 under 35 U.S.C. 101 has been withdrawn.
In particular, upon further consideration in view of the newly made amendments the claims are successfully integrated into a practical application. Specifically, the claims improve the ability of a machine learning models to process CAD models. The irregular data structure of B-reps, an extremely common 3D data format, has historically made it extremely difficult to process said data with conventional machine learning models. The generation of graphs and UV grids from the B-rep model, as claimed, allows machine learning models to consistently and efficiently process B-rep data, something that had not been possible before.
Response to Arguments -103
Applicant's arguments filed 0 have been fully considered but they are not persuasive.
Applicant argues that, based on the previous mapping, Blender would have had to include a structure that maps a B-rep to a set of grids, to which it is allegedly silent.
Examiner responds by explaining that Blender alone is not relied upon to teach this feature, rather the combination of Kim, Cao, Yang, and Blender teaches this, with the OBJ files of Blender being relied upon as a UV data container and Kim, Cao, and Yang teaching the particulars of said contained data.
Applicant argues that no prior art teaches the newly amended limitations included in the independent claims, specifically “computing a final result via a trained graph neural network based on the set of node feature vectors, a first graph included in the first UV-net representation, and a set of edge feature vectors based on the first UV-net representation, wherein the first UV-net representation comprises a data structure that maps the B-rep of the first 3D CAD object to a set of grids, and wherein the first trained neural network generates results for the first 3D CAD object independent of a B-rep structure used to represent the first 3D CAD object.”
Examiner responds by explaining that these limitations are taught by the previously cited references.
Particularly, Kim teaches computing a final result ([Col 5 line 27-31] “ In particular, a neural network can include a machine-learning model that utilizes algorithms to learn from, and make predictions on, known data by analyzing the known data to learn to generate outputs that reflect patterns and attributes of the known data.” [Col 10 line 21-25] “ In one or more embodiments, the cycle projection system 102 predicts the coordinate in the surface mapping space by using the first image as input to the pixel mapping neural network, which can then output a unit vector per pixel representing a point on the surface of a sphere.”) ([Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.”) ([Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.”)
Cao makes obvious computing a result via a trained graph neural network based on the set of node feature vectors, a first graph included in the first ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face” [Page 6 Col 2 Par 6] “CAD shapes generated by the method in Section 2 are first converted into graph representations described in Section 3.1, and then used as input to the graph neural network described in Section 3.2.”) ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face” [Page 6 Col 2 Par 6] “CAD shapes generated by the method in Section 2 are first converted into graph representations described in Section 3.1, and then used as input to the graph neural network described in Section 3.2.”) wherein the first ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face” [Page 6 Col 2 Par 6] “CAD shapes generated by the method in Section 2 are first converted into graph representations described in Section 3.1, and then used as input to the graph neural network described in Section 3.2.” [Page 8 Col 2 Par 2-6] “Before training, the CAD models are converted into graphs and point clouds. Mappings between B-Rep faces and individual points in point clouds are stored so that the results can be propagated from points back to faces. … The 3D CAD shape was converted into point clouds using the following two steps. First, it is discretized into triangulations. Second, for each triangle, points are sampled in parametric UV space, then converted to cartesian space. There are three steps to sample the UV points...” [Examiner’s note: the process of Page 8 Col 2 Par 2-6 describes mapping the B-rep data to UV-space. By discretizing the model into triangles and sampling each triangle in UV-space, this creates a UV grid for each triangle]) and wherein the trained neural network generates results for the first 3D CAD object independent of a B-rep structure used to represent the first 3D CAD object. ([Page 8 Col 2 Par 2-6] “Before training, the CAD models are converted into graphs and point clouds. Mappings between B-Rep faces and individual points in point clouds are stored so that the results can be propagated from points back to faces. … The 3D CAD shape was converted into point clouds using the following two steps. First, it is discretized into triangulations. Second, for each triangle, points are sampled in parametric UV space, then converted to cartesian space. There are three steps to sample the UV points: 1. Points are evenly sampled along the longest edge. The number of points to be sampled is determined by dividing the length (in cartesian space) of the edge by a resolution. The resolution specifies the maximum distance (in cartesian space) between two sample points. 2. For each sample point on the longest edge, line segments are extracted, starting from the point and parallel to the shortest edge. 3. Points are sampled from each line segment. The number of points to be sampled from the shortest edge is determined in the same way as in Step 1. For other line segments, the number of points is interpolated linearly between 1 and the number of points on the shortest edge, according to the distance of the line segment to the shortest edge.” [Page 9 Col 1 Par 4] “During testing, the predicted results of SSCN are transferred back to boundary faces of the CAD shapes using a voting scheme and written to files. For each face, the sample points vote for their predicted labels. The label which wins the most points is assigned to the face. Face labels predicted by the GNN are also recorded.”)
[Page 8 Col 2 Par 2-6] and subsequent [Page 9 Col 1 Par 4] of Cao describe mapping the B-rep data to UV space, then using this converted data to generate prediction results using the graph neural network.
Note that in view of the specification, the results being generated “independent of a B-rep structure” merely means that the results are not pulled directly from the B-rep structure, i.e. the structure is first converted to some other form and results are generated from that converted form. This is what is described in the specification, which is equivalent to what is disclosed in Cao. See [Par 124] of the specification “In some embodiments, a training application trains a machine learning model to map UV-net representations to final results. Each final result includes, without limitation, a set of face results for the set of nodes in the UV-net representation, a shape result, or both. In operation, the training application converts B-reps included in a B-rep training database to UV-net representations via the parameter domain graph application. Based on the UV- net representations, the training application executes any number and/or types of machine learning algorithms to train the end-to-end machine learning model to generate a trained end-to-end machine learning model. For instance, In some embodiments, the training application implements an end-to-end supervised learning process based on any number and/or types of labels associated with the B-reps.” as well as [Par 126] “For a given B-rep, the inference application converts the B-rep to a UV-net representation via the parameter domain graph application. The training application inputs the UV-net representation into the trained end-to-end machine learning model. In response, the trained end-to-end machine learning model outputs a final result corresponding to the 3D CAD object originally represented by the B- rep. In some embodiments, the final result includes, without limitation, a set of node embeddings, a shape embedding, a set of face results, a shape result, or any combination thereof.”) This process, which the claim language attempts to capture, is clearly not a process that is wholly unrelated to the B-rep structure, but is rather a process that first converts the B-rep structure to another form before doing final processing on that new form.
Further note that the way the mapping has been presented has been altered slightly for clarity but the mapping itself is the same as in the previous action. For example, the previous rejection recited “…and a first graph included in ([Page 7 Par 3] “ In this paper, …” [Page 6 Col 2 Par 6] “CAD shapes …”) while the present rejection presents this as “and a first graph included in the first ([Page 7 Par 3] “ In this paper, …” [Page 6 Col 2 Par 6] “CAD shapes …”). In both cases the passages of Cao are mapped to “a first graph included in the first representation” with a later reference teaching that such a representation is a “UV-net representation”.
Yang makes obvious ([Page 597 Par 3] “ In this section, we propose a novel model to incorporate Node and Edge features in graph Neural Networks (NENN) based on a hierarchical dual-level attention mechanism. For a graph G(V, E) with node features and edge features, where V defines a set of Nv = |V | nodes, E is a set of Ne = |E| edges. Let X = {xi |i ∈ Nv} ∈ R Nv×dv be node feature matrix, where xi ∈ R dv represents dv-dimensional feature vector of node i. Let E = {ei |i ∈ Ne} ∈ R Ne×de be edge feature matrix, where ei ∈ R de denotes de-dimensional feature vector of edge i.”)
Blender makes obvious a first UV-net representation; data based on the first UV-net representation, and wherein the first UV-net representation comprises data. ([Page 1 Par 1] “OBJ is a widely used de facto standard in the 3D industry. The OBJ format is a popular plain text format, however, it has only basic geometry and material support.
Mesh: vertices, faces, edges, normals, UV’s
Separation by groups/objects Materials/textures
NURBS curves and surfaces”
[Examiner’s note: The OBJ file format bundles CAD model geometry, UV maps and grids, textures/materials and BREP information (Note that NURBS is a type of BREP) into a single container/representation])
Applicant argues that no prior art teaches the content of new claim 21, particularly “wherein computing the final result comprises generating a probability vector for each node included in the first graph to identify a feature label for each face in the B-rep of the first 3D CAD object.”
Examiner responds by explaining that this is clearly taught by previously cited reference Cao. In particular, Cao teaches wherein computing the final result comprises generating a probability vector for each node included in the first graph to identify a feature label for each face in the B-rep of the first 3D CAD object. ([Page 8 Col 2 Par 2-6] “Before training, the CAD models are converted into graphs and point clouds. Mappings between B-Rep faces and individual points in point clouds are stored so that the results can be propagated from points back to faces. … The 3D CAD shape was converted into point clouds using the following two steps. First, it is discretized into triangulations. Second, for each triangle, points are sampled in parametric UV space, then converted to cartesian space. There are three steps to sample the UV points: …” [Page 7 Col 2 Par 8] “The adopted graph neural network architecture follows the Pytorch Geometric [22] implementation of dynamic graph CNN [16] for segmentation. Figure 8 shows the architecture of the GNN. The input is a graph as described in Section 3.1, where the number of nodes may vary and each node has a feature vector of length 4. GNN performs convolutions on the graph edges as described in [14]. After three edge convolutions, GNN predicts a probability vector of length 𝐹 for each node as outputs, where F−1 is the number of feature types. The output vectors are compared with ground truth labels to compute a negative log likelihood loss.” [Page 8 Col 1 Par 4] “During testing, the predicted results of SSCN are transferred back to boundary faces of the CAD shapes using a voting scheme and written to files. For each face, the sample points vote for their predicted labels. The label which wins the most points is assigned to the face. Face labels predicted by the GNN are also recorded.”)
This passage describes generating probability vectors for each node and then using these probability vectors to generate feature labels for faces of the B-rep model.
Applicant argues that no prior art teaches the content of new claim 22, particularly “wherein a first 2D UV-grid included in the plurality of 2D UV-grids specifies at least one of a 3D point position in a geometry domain, a 3D normal, or a visibility flag for each 2D grid point included in a plurality of 2D grid points in a parameter domain.”
Examiner responds by explaining that this is taught by the previously cited Kim reference. In particular, Kim makes obvious wherein a first 2D UV-grid included in the plurality of 2D UV-grids specifies at least one of a 3D point position in a geometry domain, ([Col 5 line 65- col 6 line 3] “ Moreover, a surface generation neural network can also determine a projection between coordinates of a surface mapping space (e.g., UV coordinates) and coordinates of three-dimensional mesh to identify a three-dimensional coordinate corresponding to a pixel of an object portrayed in an image” [Col 19 line 63-67] “For example, the data storage manager 410 can store information associated with images, pixel locations, mappings between coordinate spaces, three-dimensional object meshes, latent vectors, and multi-view cycle consistency losses.”) a 3D normal, or a visibility flag for each 2D grid point included in a plurality of 2D grid points in a parameter domain
The “projection between coordinates of a surface mapping space (e.g., UV coordinates) and coordinates of three-dimensional mesh” is equivalent to specifying a 3D point in the geometry domain for UV grid data.
Further note that the surface mapping space (i.e. the UV grid space) is per-surface, i.e. each surface has a separate UV grid ([Col 6 line 22-25] “As used herein, the term “surface mapping space” refers to a coordinate space that includes a two-dimensional representation of a surface of a three-dimensional object.”)
Claim Objections
Claims 1-22 objected to because of the following informalities:
1. Claims 1, 11, and 20 recite “wherein the first trained neural network generates results for the first 3D CAD object independent of a B-rep structure used to represent the first 3D CAD object.” In view of the specification it is clear that “independent” merely means that the results are not pulled directly from the B-rep structure, i.e. the structure is first converted to some other form and results are generated from that converted form. This is what is described in the specification, which is equivalent to what is disclosed in Cao. See [Par 124] of the specification “In some embodiments, a training application trains a machine learning model to map UV-net representations to final results. Each final result includes, without limitation, a set of face results for the set of nodes in the UV-net representation, a shape result, or both. In operation, the training application converts B-reps included in a B-rep training database to UV-net representations via the parameter domain graph application. Based on the UV- net representations, the training application executes any number and/or types of machine learning algorithms to train the end-to-end machine learning model to generate a trained end-to-end machine learning model. For instance, In some embodiments, the training application implements an end-to-end supervised learning process based on any number and/or types of labels associated with the B-reps.” as well as [Par 126] “For a given B-rep, the inference application converts the B-rep to a UV-net representation via the parameter domain graph application. The training application inputs the UV-net representation into the trained end-to-end machine learning model. In response, the trained end-to-end machine learning model outputs a final result corresponding to the 3D CAD object originally represented by the B- rep. In some embodiments, the final result includes, without limitation, a set of node embeddings, a shape embedding, a set of face results, a shape result, or any combination thereof.”) This process, which the claim language attempts to capture, is clearly not a process that is wholly unrelated to the B-rep structure, but is rather a process that first converts the B-rep structure to another form before doing final processing on that new form.
As this makes the most sense and matches best with the rest of the disclosure, this interpretation is used in the current examination. If, however, “independent” is instead meant to be interpreted as “wholly unconnected,” i.e. that the final result is generated wholly without information even indirectly related to the B-rep, significant issues related to written description may be present, as the specification only describes the neural network results being generated from data that is converted, mapped, extracted, and/or otherwise taken from/reliant on the B-rep data. ([Par 27] “To address the above limitations of B-reps, the system 100 includes, without limitation, a parameter domain graph application 120, a training application 130, an inference application 190, or any combination thereof. The parameter domain graph application 120 extracts topological data and geometric features from B-reps to generate UV-net representations that are amenable to processing using neural networks. The training application 130 uses the parameter domain graph application 120 to generate a trained end-to-end machine learning model 160 based on a B-rep training set 102. The trained end-to-end machine learning model 160 includes, without limitation, trained any number and/or types of neural networks that, together, map UV-net representations to final results. Each final result can include, without limitation, any amount and/or types of data that is relevant to a task associated with 3D CAD objects. The inference application 190 uses the parameter domain graph application 120 and the trained end-to-end machine learning model 160 to generate final results for 3D CAD objects represented by B-reps.”)
For clarity purposes, t is recommended to amend the limitation to instead read something along the lines of “wherein the first trained neural network generates results for the first 3D CAD object indirectly from a B-rep structure used to represent the first 3D CAD object.”
Appropriate correction is required.
Warning – Duplicate Claims
Applicant is advised that should claim 5 be found allowable, claim 22 will be objected to under 37 CFR 1.75 as being a substantial duplicate thereof. When two claims in an application are duplicates or else are so close in content that they both cover the same thing, despite a slight difference in wording, it is proper after allowing one claim to object to the other as being a substantial duplicate of the allowed claim. See MPEP § 608.01(m).
Note that claims 5 and 22 contain identical text.
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.
(1) Claims 1, 5, 9, 11, 13-14, and 19-22 are rejected under 35 U.S.C. 103 as being unpatentable over Kim (US 10937237 B1) in view of Graph Representation of 3D CAD models for Machining Feature Recognition with Deep Learning (Hereinafter Cao) in further view of NENN: Incorporate Node and Edge Features in Graph Neural Networks (Hereinafter Yang) as well as Wavefront OBJ – Blender Manual (Hereinafter Blender)
Claim 1. Kim makes obvious A computer-implemented method for performing tasks associated with three- dimensional (3D) computer-aided design (CAD) objects ([Col 2 line 49-54] “One or more embodiments of the present disclosure include a multi-view cycle projection system that utilizes neural networks to perform multi-view cycle projection across a plurality of two-dimensional images and an estimated three-dimensional object mesh of an object.” [Col 3 line 23-30] “The cycle projection system can then apply the surface generation neural network (as a decoder) to the latent vector to generate the estimated three-dimensional object mesh. The cycle projection system can generate the estimated three-dimensional object mesh and determine a corresponding location on the surface of the resulting three-dimensional object mesh from the predicted surface mapping coordinate.”) ([Col 10 line 21-30] “For example, the cycle projection system can utilize the surface generation neural network to map (u,v) coordinates (generated from the pixel mapping neural network) to the three-dimensional object mesh (determined from the latent vector representation of the object).” [Col 6 line 27-32] “For example, in one or more embodiments, a surface mapping space can include a UV space that includes (u,v) coordinates. Additionally, a surface mapping space can include a grid with values in each coordinate direction from 0 to 1 (e.g., with boundary corners at (0,0), (0,1), (1,0), and (1,1)). [Col 19 line 63-67] “For example, the data storage manager 410 can store information associated with images, pixel locations, mappings between coordinate spaces, three-dimensional object meshes, latent vectors, and multi-view cycle consistency losses.”) using a first trained neural network to generate a set of node feature vectors ([Col 5 line 46-55] “Moreover, as used herein, an “object encoder neural network” refers to a neural network that generates a latent vector representation of an object from a digital image. For example, an object encoder neural network can include a feed forward neural network that generates a feature map representation by processing an object portrayed in a digital image. As used herein, the term “latent vector” (or latent feature vector, feature map, or feature representation) refers to a feature vector of fixed length that represents a two-dimensional image. For instance, a latent vector can include a fixed length representation of one or more of the two-dimensional images of an object portrayed in a digital image.”) and computing a final result ([Col 5 line 27-31] “ In particular, a neural network can include a machine-learning model that utilizes algorithms to learn from, and make predictions on, known data by analyzing the known data to learn to generate outputs that reflect patterns and attributes of the known data.”) [Col 10 line 21-25] “ In one or more embodiments, the cycle projection system 102 predicts the coordinate in the surface mapping space by using the first image as input to the pixel mapping neural network, which can then output a unit vector per pixel representing a point on the surface of a sphere.”) ([Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.”)([Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.”)
Kim does not explicitly teach a method for performing operations related to CAD objects that are represented using boundary-representations (B-reps), a first UV-net representation, node feature vectors, and performing operations via a trained graph neural network based on the set of node feature vectors, a first graph included in the first UV-net representation, and a set of edge feature vectors based on the first UV-net representation, wherein the first UV-net representation comprises a data structure that maps the B-rep of the first 3D CAD object to a set of grids, and wherein the trained neural network generates results for the first 3D CAD object independent of a B-rep structure used to represent the first 3D CAD object.
Cao makes obvious a method for performing operations related to CAD objects that are represented using boundary-representations (B-reps), a first ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face” [Page 6 Col 2 Par 6] “CAD shapes generated by the method in Section 2 are first converted into graph representations described in Section 3.1, and then used as input to the graph neural network described in Section 3.2.”) ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face” [Page 6 Col 2 Par 6] “CAD shapes generated by the method in Section 2 are first converted into graph representations described in Section 3.1, and then used as input to the graph neural network described in Section 3.2.”) wherein the first ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face” [Page 6 Col 2 Par 6] “CAD shapes generated by the method in Section 2 are first converted into graph representations described in Section 3.1, and then used as input to the graph neural network described in Section 3.2.” [Page 8 Col 2 Par 2-6] “Before training, the CAD models are converted into graphs and point clouds. Mappings between B-Rep faces and individual points in point clouds are stored so that the results can be propagated from points back to faces. … The 3D CAD shape was converted into point clouds using the following two steps. First, it is discretized into triangulations. Second, for each triangle, points are sampled in parametric UV space, then converted to cartesian space. There are three steps to sample the UV points...” [Examiner’s note: the process of Page 8 Col 2 Par 2-6 describes mapping the B-rep data to UV-space. By discretizing the model into triangles and sampling each triangle in UV-space, this creates a UV grid for each triangle]) and wherein the trained neural network generates results for the first 3D CAD object independent of a B-rep structure used to represent the first 3D CAD object. ([Page 8 Col 2 Par 2-6] “Before training, the CAD models are converted into graphs and point clouds. Mappings between B-Rep faces and individual points in point clouds are stored so that the results can be propagated from points back to faces. … The 3D CAD shape was converted into point clouds using the following two steps. First, it is discretized into triangulations. Second, for each triangle, points are sampled in parametric UV space, then converted to cartesian space. There are three steps to sample the UV points: 1. Points are evenly sampled along the longest edge. The number of points to be sampled is determined by dividing the length (in cartesian space) of the edge by a resolution. The resolution specifies the maximum distance (in cartesian space) between two sample points. 2. For each sample point on the longest edge, line segments are extracted, starting from the point and parallel to the shortest edge. 3. Points are sampled from each line segment. The number of points to be sampled from the shortest edge is determined in the same way as in Step 1. For other line segments, the number of points is interpolated linearly between 1 and the number of points on the shortest edge, according to the distance of the line segment to the shortest edge.” [Page 9 Col 1 Par 4] “During testing, the predicted results of SSCN are transferred back to boundary faces of the CAD shapes using a voting scheme and written to files. For each face, the sample points vote for their predicted labels. The label which wins the most points is assigned to the face. Face labels predicted by the GNN are also recorded.”)
Cao is analogous art because it is within the field of CAD model processing using graph neural networks. It would have been obvious to combine it with Kim before the effective filing date. One of ordinary skill in the art would have been motivated to make this combination in order to make manufacturing designed CAD models easier by automating recognition of different machining requirements of different parts ([Page 1 Col 2 Par 2 – Page 2 Col 1 Par 2] “Considering an engineering component, its shape can be thought of as consisting of engineering features (e.g. holes, bosses and chamfers). Each of these features may require different treatment in CAE and CAM tools, and the ability to identify engineering features would allow many CAE/CAM process to be automated. This would allow, for example, the ability to automatically tailor the mesh and boundary conditions (in CAE terms) or to automate manufacturing simulations to help improve the overall process efficiency. However, engineering features are usually not explicitly defined in CAD models… Therefore, it would be extremely valuable in engineering processes to be able to automatically identify machining features from CAD shapes…. Therefore, the focus of this paper is to show how CAD/CAE/CAM industries could benefit from deep learning methodologies, and more specifically their application to machining feature recognition.”) One of ordinary skill in the art would have recognized that combining Cao with Kim would result in a system that was better able to recognize design features and use that recognition to enable easier, more efficient manufacturing practices.
The combination of Kim and Cao does not explicitly teach the first UV-net representation, and a set of edge feature vectors based on the first UV-net representation, wherein the first UV-net representation comprises data.
Yang makes obvious ([Page 597 Par 3] “ In this section, we propose a novel model to incorporate Node and Edge features in graph Neural Networks (NENN) based on a hierarchical dual-level attention mechanism. For a graph G(V, E) with node features and edge features, where V defines a set of Nv = |V | nodes, E is a set of Ne = |E| edges. Let X = {xi |i ∈ Nv} ∈ R Nv×dv be node feature matrix, where xi ∈ R dv represents dv-dimensional feature vector of node i. Let E = {ei |i ∈ Ne} ∈ R Ne×de be edge feature matrix, where ei ∈ R de denotes de-dimensional feature vector of edge i.”)
Yang is analogous art because it is within the field of graph neural network operations. It would have been obvious to combine it with Kim and Cao before the effective filing date. One of ordinary kill in the art would have been motivated to make this combination in order to take better advantage of the connected nature of graphs. As stated by Yang, previous systems only took into account features associated with individual nodes, which makes it hard to measure information relating to entire neighborhoods of nodes ([Page 594 Par 2] “ Despite the success of existing graph neural networks, there are two enormous challenges. On the one hand, almost all previous literatures only leverage the node features and completely ignore the edge features that are completely likely to contain important information. For example, in molecule networks, a node represents an atom while an edge represents a bond connecting two atoms. A bond usually has some simple edge features (e.g., bond type, atom pair type, bond order, conjugated, ring status, aromaticity), which are closely related to atom features. On the other hand, how to measure the importance of neighborhood as well as the connecting edges or nodes is not fully considered”) To this end, Yang introduces a new type of graph neural network system that takes into account both the features of nodes and of the edges that connect them ([Page 954 Par 3] “ In order to address the aforementioned challenges, we propose a novel graph neural network, named NENN, which incorporates node and edge features based on a dual-level attention mechanism, including node-level and edge-level attentions. Specifically, we aim to to learn the importance of node based neighbors and edge based neighbors and aggregate embeddings for each node in the node-level attention layer. Similarly, the embedding of each edge is generated in the edge-level attention layer.”) Overall, one of ordinary skill in the art would have recognized that combining Kim and Cao with Yang would result in a more robust graph neural network system capable of more accurately processing graph data.
The combination of Kim, Cao, and Yang does not explicitly teach a first UV-net representation; data based on the first UV-net representation, and wherein the first UV-net representation comprises data.
Blender makes obvious a first UV-net representation; data based on the first UV-net representation, and wherein the first UV-net representation comprises data. ([Page 1 Par 1] “OBJ is a widely used de facto standard in the 3D industry. The OBJ format is a popular plain text format, however, it has only basic geometry and material support.
Mesh: vertices, faces, edges, normals, UV’s
Separation by groups/objects Materials/textures
NURBS curves and surfaces”
[Examiner’s note: The OBH file format bundles CAD model geometry, UV maps and grids, textures/materials and BREP information (Note that NURBS is a type of BREP) into a single container/representation])
Blender is analogous art because it is within the field of 3D CAD. It would have been obvious to one of ordinary skill in the art to combine it with Kim, Cao, and Yang before the effective filing date. One of ordinary skill in the art would have been motivated to make this combination in order to make the system compatible with industry standard CAD models, enabling easier, more widespread use. As stated by Blender, ([Page 1 Par 1]“OBJ is a widely used de facto standard in the 3D industry.”) Blender also allows the import/export of OBJ files ([Page 1 Par 4] “Usage Import/Export geometry and curves to the OBJ format.”) One of ordinary skill in the art would have recognized that enabling the system to import/export .OBJ files by combining Blender with Kim, Cao, and Yang would allow deeper integration with other modelling systems a user may already be using and allow easy transition between systems.
Claim 11. The elements of claim 11 are substantially the same as those of claim 1. Therefore, the elements of claim 11 are rejected due to the same reasons as outlined above for claim 1. Further, Kim makes obvious the additional elements of “One or more non-transitory computer readable media including instructions that, when executed by one or more processors, cause the one or more processors to perform tasks” ([Col 20 Line 9-13] “Alternatively, a non-transitory computer readable medium can comprise instructions, that when executed by one or more processors, cause a computing device to perform the acts”).
Claim 20. The elements of claim 20 are substantially the same as those of claim 1. Therefore, the elements of claim 20 are rejected due to the same reasons as outlined above for claim 1. Further, Kim makes obvious the additional elements of “one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:” ([Col 23 Line 49-56] “In one or more embodiments, the processor 602 includes hardware for executing instructions, such as those making up a computer program. As an example, and not by way of limitation, to execute instructions for dynamically modifying workflows, the processor 602 may retrieve (or fetch) the instructions from an internal register, an internal cache, the memory 604, or the storage device 606 and decode and execute them.”).
Claim 5. Kim makes obvious wherein a first 2D UV-grid included in the plurality of 2D UV-grids specifies at least one of a 3D point position in a geometry domain, ([Col 5 line 65- col 6 line 3] “ Moreover, a surface generation neural network can also determine a projection between coordinates of a surface mapping space (e.g., UV coordinates) and coordinates of three-dimensional mesh to identify a three-dimensional coordinate corresponding to a pixel of an object portrayed in an image” [Col 19 line 63-67] “For example, the data storage manager 410 can store information associated with images, pixel locations, mappings between coordinate spaces, three-dimensional object meshes, latent vectors, and multi-view cycle consistency losses.”) a 3D normal, or a visibility flag for each 2D grid point included in a plurality of 2D grid points in a parameter domain
Claim 14. The elements of claim 14 are substantially the same as those of claim 5. Therefore, the elements of claim 14 are rejected due to the same reasons as outlined above for claim 5. Further, Kim makes obvious the additional elements of claim 11 from which claim 14 descends, particularly “one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:” ([Col 23 Line 49-56] “In one or more embodiments, the processor 602 includes hardware for executing instructions, such as those making up a computer program. As an example, and not by way of limitation, to execute instructions for dynamically modifying workflows, the processor 602 may retrieve (or fetch) the instructions from an internal register, an internal cache, the memory 604, or the storage device 606 and decode and execute them.”).
Claim 9. Blender makes obvious further comprising computing the first UV-net representation based on a B-rep of the first 3D CAD object. ([Page 3 Par 9] “Export – NURBS Write out NURBS curves as OBJ NURBS rather than converting to geometry”[Examiner’s note: creating an OBJ file from a blender design involves processing the NURBS curves of the design, NURBS being type of BREP])
Claim 19. The elements of claim 19 are substantially the same as those of claim 9. Therefore, the elements of claim 19 are rejected due to the same reasons as outlined above for claim 9. Further, Kim makes obvious the additional elements of claim 11 from which claim 19 descends, particularly “one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:” ([Col 23 Line 49-56] “In one or more embodiments, the processor 602 includes hardware for executing instructions, such as those making up a computer program. As an example, and not by way of limitation, to execute instructions for dynamically modifying workflows, the processor 602 may retrieve (or fetch) the instructions from an internal register, an internal cache, the memory 604, or the storage device 606 and decode and execute them.”).
Claim 13. Kim makes obvious wherein the first trained neural network comprises ([Col 5 line 46-55] “Moreover, as used herein, an “object encoder neural network” refers to a neural network that generates a latent vector representation of an object from a digital image. For example, an object encoder neural network can include a feed forward neural network that generates a feature map representation by processing an object portrayed in a digital image.”) ([Col 10 line 64 – col 11 line 4] “ In other words, the surface generation neural network can transform one or more 2D surfaces (e.g., in a UV space) to a surface, … The parameters of the transformations come both from the learned weights of the neural network and the learned representation of the shape.” [Col 13 line 2-6] “To illustrate, the cycle projection system 102 can utilize back-propagation techniques to modify one or more weights or values of the pixel mapping neural network in a way that reduces the multi-view cycle consistency loss.” ) [Col 6 line 27-32] “For example, in one or more embodiments, a surface mapping space can include a UV space that includes (u,v) coordinates. Additionally, a surface mapping space can include a grid with values in each coordinate direction from 0 to 1 (e.g., with boundary corners at (0,0), (0,1), (1,0), and (1,1)).” [Col 19 line 63-67] “For example, the data storage manager 410 can store information associated with images, pixel locations, mappings between coordinate spaces, three-dimensional object meshes, latent vectors, and multi-view cycle consistency losses.”)
Cao makes obvious a convolution neural network ([Page 7 Col 2 Par 8] “ The adopted graph neural network architecture follows the Pytorch Geometric [22] implementation of dynamic graph CNN [16] for segmentation.”)
Claim 21. Cao teaches wherein computing the final result comprises generating a probability vector for each node included in the first graph to identify a feature label for each face in the B-rep of the first 3D CAD object. ([Page 8 Col 2 Par 2-6] “Before training, the CAD models are converted into graphs and point clouds. Mappings between B-Rep faces and individual points in point clouds are stored so that the results can be propagated from points back to faces. … The 3D CAD shape was converted into point clouds using the following two steps. First, it is discretized into triangulations. Second, for each triangle, points are sampled in parametric UV space, then converted to cartesian space. There are three steps to sample the UV points: …” [Page 7 Col 2 Par 8] “The adopted graph neural network architecture follows the Pytorch Geometric [22] implementation of dynamic graph CNN [16] for segmentation. Figure 8 shows the architecture of the GNN. The input is a graph as described in Section 3.1, where the number of nodes may vary and each node has a feature vector of length 4. GNN performs convolutions on the graph edges as described in [14]. After three edge convolutions, GNN predicts a probability vector of length 𝐹 for each node as outputs, where F−1 is the number of feature types. The output vectors are compared with ground truth labels to compute a negative log likelihood loss.” [Page 8 Col 1 Par 4] “During testing, the predicted results of SSCN are transferred back to boundary faces of the CAD shapes using a voting scheme and written to files. For each face, the sample points vote for their predicted labels. The label which wins the most points is assigned to the face. Face labels predicted by the GNN are also recorded.”)
Claim 22. Kim teaches wherein a first 2D UV-grid included in the plurality of 2D UV-grids specifies at least one of a 3D point position in a geometry domain, ([Col 5 line 65- col 6 line 3] “ Moreover, a surface generation neural network can also determine a projection between coordinates of a surface mapping space (e.g., UV coordinates) and coordinates of three-dimensional mesh to identify a three-dimensional coordinate corresponding to a pixel of an object portrayed in an image” [Col 19 line 63-67] “For example, the data storage manager 410 can store information associated with images, pixel locations, mappings between coordinate spaces, three-dimensional object meshes, latent vectors, and multi-view cycle consistency losses.”) a 3D normal, or a visibility flag for each 2D grid point included in a plurality of 2D grid points in a parameter domain.
(2) Claims 8 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Kim (US 10937237 B1) in view of Graph Representation of 3D CAD models for Machining Feature Recognition with Deep Learning (Hereinafter Cao) in further view of NENN: Incorporate Node and Edge Features in Graph Neural Networks (Hereinafter Yang) as well as Wavefront OBJ – Blender Manual (Hereinafter Blender) and 3D Object Classification Using a Volumetric Deep Neural Network: An Efficient Octree Guided Auxiliary Learning Approach (Hereinafter Muzahid)
Claim 8. Kim makes obvious processing [Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.”) to generate a shape embedding ([Col 10 line 60-64] “ In particular, the surface generation neural network can approximate a target surface by mapping one or more 2D shapes (e.g., a set of squares) to the surface of a 3D shape. The surface generation neural network can jointly learn a parameterization and an embedding of the shape.” and ([Col 10 line 60-64] “ In particular, the surface generation neural network can approximate a target surface by mapping one or more 2D shapes (e.g., a set of squares) to the surface of a 3D shape. The surface generation neural network can jointly learn a parameterization and an embedding of the shape.”) using a trained non-linear classifier ([Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.” [Examiner’s note: a multi-layer perceptron is a type of nonlinear classifier]) to ([Col 3 line 23-30] “The cycle projection system can then apply the surface generation neural network (as a decoder) to the latent vector to generate the estimated three-dimensional object mesh. The cycle projection system can generate the estimated three-dimensional object mesh and determine a corresponding location on the surface of the resulting three-dimensional object mesh from the predicted surface mapping coordinate.”)
Cao makes obvious the first graph, set of node feature vectors, and ([Page 7 Par 3] “ In this paper, a graph representation constructed from the B-Rep model is proposed as input for deep neural networks. Given the B-Rep model of a CAD shape, the graph representation is constructed following three steps: 1. A node is created for each face. 2. A link between two nodes is created if faces associated with these nodes are adjacent (i.e. if they share an edge). 3. Each node is assigned a feature vector representing the geometry of the face”)
Yang makes obvious a set of edge feature vectors; ([Page 597 Par 3] “ In this section, we propose a novel model to incorporate Node and Edge features in graph Neural Networks (NENN) based on a hierarchical dual-level attention mechanism. For a graph G(V, E) with node features and edge features, where V defines a set of Nv = |V | nodes, E is a set of Ne = |E| edges. Let X = {xi |i ∈ Nv} ∈ R Nv×dv be node feature matrix, where xi ∈ R dv represents dv-dimensional feature vector of node i. Let E = {ei |i ∈ Ne} ∈ R Ne×de be edge feature matrix, where ei ∈ R de denotes de-dimensional feature vector of edge i.”)
The combination of Kim, Cao, Yang, and Blender fails to make obvious processing data to generate a predicted classification for the 3D CAD object.
Muzahid makes obvious processing data to generate a predicted classification for the 3D CAD object ([Abstract] “In this context, we propose a multiorientation volumetric deep neural network (MV-DNN) for 3D object classification with octree generating low-cost volumetric features.” [Fig. 1] shows 3D CAD object classification based on extracted features)
PNG
media_image1.png
217
682
media_image1.png
Greyscale
Muzahid is analogous art because it is within the field of CAD model neural network processing. It would have been obvious to one of ordinary skill in the art to combine it with Kim, Cao, Yang, and Blender before the effective filing date. One of ordinary skill in the art would have been motivated to make this combination to create a more efficient CAD neural network processing system. ([Abstract] “We consider the recent challenges of 3D shape analysis based on a volumetric CNN that requires a huge computational power. This high-cost approach forces to reduce the volume resolutions when applying 3D CNN on volumetric data. In this context, we propose a multiorientation volumetric deep neural network (MV-DNN) for 3D object classification with octree generating low-cost volumetric features. In comparison to conventional octree representations, we propose to limit the octree partition to a certain depth to reserve all leaf octants with sparsity features. This allows for improved learning of complex 3D features and increased prediction of object labels at both low and high resolutions…. Our deep learning framework runs significantly faster and consumes less memory than full voxel representations and demonstrate the effectiveness of our octree-based auxiliary learning approach for exploring high resolution 3D models. “) Overall, one of ordinary skill in the art would have recognize that combining Kim, Cao, Yang, and Blender with Muzahid would result in a model processing system that is significantly more efficient.
Claim 18. The elements of claim 18 are substantially the same as those of claim 8. Therefore, the elements of claim 18 are rejected due to the same reasons as outlined above for claim 8. Further, Kim makes obvious the additional elements of claim 11 from which claim 18 descends, particularly “one or more memories storing instructions; and one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:” ([Col 23 Line 49-56] “In one or more embodiments, the processor 602 includes hardware for executing instructions, such as those making up a computer program. As an example, and not by way of limitation, to execute instructions for dynamically modifying workflows, the processor 602 may retrieve (or fetch) the instructions from an internal register, an internal cache, the memory 604, or the storage device 606 and decode and execute them.”).
(3) Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Kim (US 10937237 B1) in view of Graph Representation of 3D CAD models for Machining Feature Recognition with Deep Learning (Hereinafter Cao) in further view of NENN: Incorporate Node and Edge Features in Graph Neural Networks (Hereinafter Yang) as well as Wavefront OBJ – Blender Manual (Hereinafter Blender) (Hereinafter Blender) as well as On Visual Similarity Based 3D Model Retrieval (Hereinafter Chen)
Claim 10. Kim makes obvious wherein the final result ([Col 5 line 27-31] “ In particular, a neural network can include a machine-learning model that utilizes algorithms to learn from, and make predictions on, known data by analyzing the known data to learn to generate outputs that reflect patterns and attributes of the known data.”) specifies at least one of a set of node embeddings or a shape embedding, and ([Col 10 line 63-64] “ The surface generation neural network can jointly learn a parameterization and an embedding of the shape.”) ([Col 5 line 27-31] “…to generate outputs that reflect patterns and attributes of the known data.”) ([Col 5 line 27-31] “…to generate outputs that reflect patterns and attributes of the known data.”)
Kim fails to make obvious further comprising transmitting input to a CAD tool that determines one or more similarities in shape between the first 3D CAD object and a second 3D CAD object
Chen makes obvious further comprising transmitting input ([Page 5 Col 2 Par 2] “ ) Translation and scaling are applied first to ensure that 3D model is entirely contained in rendered images. The input 3D model is translated from the center of the model to the origin of world coordinate system. The axis is then scaled such that the maximum length is 1.”) to a CAD tool that determines one or more similarities in shape between the first 3D CAD object and a second 3D CAD object ([Page 1 col 1 Par 1] “In this paper, however, we present a novel approach that matches 3D models using their visual similarities, which are measured with image differences in light fields” [Page 6 Col 1 Par 2] “The retrieving process can be referred as calculating the similarity one by one between the queried one and each of the models in the database and showing those similar to the queried one. The similarity between two models is defined as summing up the similarity from all the corresponding images, as described in Section 2.3.”)
Chen is analogous art because it is within the field of 3D CAD model processing. It would have been obvious to one of ordinary skill in the art to combine it with Kim, Cao, Yang, and Blender before the effective filing date. One of ordinary skill in the art would have been motivated to make this combination in order to more efficiently process and compare models. ([Abstract] “ In this paper, a visual similarity-based 3D model retrieval system is proposed. This approach measures the similarity among 3D models by visual similarity, and the main idea is that if two 3D models are similar, they also look similar from all viewing angles … The visual similarity-based approach is robust against similarity transformation, noise, model degeneracy etc., and provides 42%, 94% and 25% better performance (precision-recall evaluation diagram) than three other competing approaches:”) One of ordinary skill in the art would have recognized that combining Chen with Kim, Cao, Yang, and Blender would result in a system capable of processing and comparing models and associated embeddings much more efficiently.
(4) Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Kim (US 10937237 B1) in view of Graph Representation of 3D CAD models for Machining Feature Recognition with Deep Learning (Hereinafter Cao) in further view of NENN: Incorporate Node and Edge Features in Graph Neural Networks (Hereinafter Yang) as well as Wavefront OBJ – Blender Manual (Hereinafter Blender) as well as Graph Representation Learning – Chapter 5: The Graph Neural Network Model (Hereinafter Hamilton)
Claim 12. Kim makes obvious wherein computing the final result ([Col 5 line 27-31] “ In particular, a neural network can include a machine-learning model that utilizes algorithms to learn from, and make predictions on, known data by analyzing the known data to learn to generate outputs that reflect patterns and attributes of the known data.”) comprises causing the trained graph neural network ([Col 5 line 32-36] “ For instance, a neural network can include, but is not limited to, a convolutional neural network, a recurrent neural network, a generative adversarial neural network, a variational auto-encoder, a feed forward neural network, a multi-layer perceptron, or a graph neural network.”)
Kim fails to make obvious causing the graph neural network to execute a message passing algorithm that propagates the set of node feature vectors over the first graph to generate a set of node embeddings.
Hamilton makes obvious causing the graph neural network to execute a message passing algorithm that propagates the set of node feature vectors over the first graph to generate a set of node embeddings. ([Page 48 Par 3-4] “The basic graph neural network (GNN) model can be motivated in a variety of ways. The same fundamental GNN model has been derived as a generalization of convolutions to non-Euclidean data [Bruna et al., 2014], as a differentiable variant of belief propagation [Dai et al., 2016], as well as by analogy to classic graph isomorphism tests [Hamilton et al., 2017b]. Regardless of the motivation, the defining feature of a GNN is that it uses a form of neural message passing in which vector messages are exchanged between nodes and updated using neural networks [Gilmer et al., 2017]…. We will focus on the message passing framework itself and defer discussions of training and optimizing GNN models to Chapter 6. The bulk of this chapter will describe how we can take an input graph G = (V, E), along with a set of node features X 2 Rd⇥|V|, and use this information to generate node embeddings zu, 8u 2 V” [Figure 5.1] “Overview of how a single node aggregates messages from its local neighborhood. The model aggregates messages from A’s local graph neighbors (i.e., B, C, and D), and in turn, the messages coming from these neighbors are based on information aggregated from their respective neighborhoods, and so on.”)
Hamilton is analogous art because it is within the field of graph neural network operations. It would have been obvious to combine it with Kim, Cao, Yang, and Blender before the effective filing date. One of ordinary skill in the art would have been motivated to make this combination in order to better apply deep learning to graph-structured data ([Page 47 Par 2] “ The primary challenge in developing complex encoders for graph-structured data is that our usual deep learning toolbox does not apply. For example, convolutional neural networks (CNNs) are well-defined only over grid-structured inputs (e.g., images), while recurrent neural networks (RNNs) are well-defined only over sequences (e.g., text). To define a deep neural network over general graphs, we need to define a new kind of deep learning architecture.“ [Page 47 Par 3] “One reasonable idea for defining a deep neural network over graphs would be to simply use the adjacency matrix as input to a deep neural network. For example, to generate an embedding of an entire graph we could simply flatten the adjacency matrix and feed the result to a multi-layer perceptron (MLP): ”) [Page 50 Par 1] “ The basic intuition behind the GNN message-passing framework is straightforward: at each iteration, every node aggregates information from its local neighborhood, and as these iterations progress each node embedding contains more and more information from further reaches of the graph. To be precise: after the first iteration (k = 1), every node embedding contains information from its 1-hop neighborhood, i.e., every node embedding contains information about the features of its immediate graph neighbors”) Overall, one of ordinary skill in the art would have recognized that combining Kim, Cao, Yang, and Blender with Hamilton would result in a system better capable of handling graph-structured data.
Allowable Subject Matter
Additionally, Claims 2-4, 6-7, and 15-17 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims in a way that overcomes the previously outlined rejection under 35 U.S.C. 101.
Claims 2-4 would be allowed under 35 U.S.C. 103 over prior art. The following is a statement of reasons for the indication of allowable subject matter: prior art representative of the claim, in particular the processing of one-dimensional UV grids included in UV-net structure could not be found.
Claims 6 and 15-16 would be allowed under 35 U.S.C. 103 over prior art. The following is a statement of reasons for the indication of allowable subject matter: prior art representative of the claim, in particular the recursive processing of specifically hidden node and edge feature vectors could not be found.
Claims 7 and 17 would be allowed under 35 U.S.C. 103 over prior art. The following is a statement of reasons for the indication of allowable subject matter: prior art representative of the claim, in particular concatenating shape embeddings to each node embedding could not be found.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Michael P Mirabito whose telephone number is (703)756-1494. The examiner can normally be reached M-F 10:30 am - 6:30 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Emerson Puente can be reached at (571) 272-3652. 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.P.M./ Examiner, Art Unit 2187
/EMERSON C PUENTE/ Supervisory Patent Examiner, Art Unit 2187
Prosecution Insights