Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
Arguments and amendments filed 2/13/2026 have been examined.
Claims 1-16 are cancelled.
Claims 17-36 have been added.
Thus, Claims 17-36 are currently pending.
This Office Action is Final.
Response to Arguments
As to the argument: “This rejection is moot in view of the cancellation of claims 1-16, and Applicant respectfully requests withdrawal of the 35 U.S.C. § 112(b) rejection of claims 1-16.”; the rejection under 35 USC 112(b) is withdrawn.
Applicant’s arguments with respect to claim(s) regarding rejection under 35 USC 103 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
As Applicant has cancelled claims 1-16 the rejection under 35 USC 101 has been withdrawn for those claims.
Applicant's arguments filed regarding rejections under 35 USC 101 and newly added claims 17-36 have been fully considered but they are not persuasive.
As to the argument:
“Applicant notes that newly-added claim 17 (and, similarly, claim 27) recites "provide the
first pooling matrix as input to an artificial intelligence model to output a prediction result relating
to the input data", which provides the requisite integration of the judicial exception into a
practical application to overcome this rejection.”
The Examiner respectfully disagrees.
Please note the relevant and recent decision, Recentive Analytics, Inc. v. Fox Corp.
Where the Federal Circuit issued a precedential decision in Recentive Analytics
(decision available via https://www.cafc.uscourts.gov/opinions-orders/23-2437.OPINION.4-18-
2025_2500790.pdf ) invalidating generic machine learning claims under Section 101 via the
two-step Alice test patents using machine learning models. The Federal Circuit acknowledged
that “[m]achine learning is a burgeoning and increasingly important field and may lead to patenteligible improvements in technology.” Nevertheless, the Federal Circuit held that “patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under § 101.”
In this case, the application of machine learning (here, a generic “a large language model
trained on a text corpus”) is completely generic and unspecific, and thus as recited in Recentive
Analytics, the claimed machine learning limitations i.e. the “artificial intelligence model” of the current claims “do no more than claim the application of generic machine learning to new data
environments”; thus the above argument is moot and the rejection under 35 USC 101 remains.
As to the argument:
“Based on these paragraphs, new dependent claims 22 and 32 have been added that
additionally recite: combine the grouping matrix and the first pooling matrix to obtain, through a
Feed-Forward Network, an output graph in which one group operates as one node, the
output graph reflecting the hierarchical structure of the input graph data, regardless of
the numbering order of the input graph data; ....
Accordingly, Applicant respectfully submits that the 35 U.S.C. § 101 rejection is moot
and should be withdrawn with regard to canceled claims 1-17, and that it should not be applied
to new claims 17-36.”
The Examiner respectfully disagrees.
As Applicant has added new claims, these arguments do not reflect the current rejection under 35 USC 101. For example, nowhere above do Applicant’s arguments address the current 101 rejection which deals with the limitation “automatically determine, based on the input data, a number of clusters for performing pooling;”, where Applicant’s assertions regarding “grouping matrix” and “Feed-forward Network” above do not address the generic and unspecified steps of “automatically determine, based on the input data, a number of clusters for performing
pooling;”. As Applicant has not specifically addressed the rejection under 35 USC 101 and the specific limitations therein, the Examiner is not convinced and the rejection under 35 USC 101 remains.
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.
Claims 17 and 27 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 17 and 27 recite:
“provide the first pooling matrix as input to an artificial intelligence model to output a
prediction result relating to the input data”/” providing the first pooling matrix as input to an artificial intelligence model to output a prediction result relating to the input data”; however, the examiner searched the relevant priority documents/specification for support of the above limitations and could not find support for “provide the first pooling matrix as input to an artificial intelligence model to output a prediction result relating to the input data”.
The Specification, for example, recites (at para. [0058]):
PNG
media_image1.png
116
730
media_image1.png
Greyscale
However, while the specification discusses “assume an inductive graph-level prediction setting” by “learning a function” (see above), nowhere does the specification provide support for the claimed limitation “provide the first pooling matrix as input to an artificial intelligence model to output a prediction result relating to the input data”; thus the claims must be rejected under 35 USC 112(a).
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 17-36 are rejected under 35 U.S.C. 101 because the claimed invention is
directed to an abstract idea without significantly more.
Claim 17 recites: (Step 2a, Prong One)
determining a number of clusters for performing pooling based on the input data.
The limitation of determining a number of clusters for performing pooling based on the input data, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, other than reciting a generic processor/memory, nothing in the claim element precludes the step from practically being performed in the mind. For example, but for the processor language, “determining” in the context of this claim encompasses the user manually determining a number of generic “clusters” using generic “input data” steps. Similarly, the limitation(s) of
acquiring; generating and providing, as drafted, is a process that, under its broadest reasonable
interpretation, covers performance of the limitation in the mind but for the recitation of generic
computer components. For example, but for the processor language, acquiring; generating and providing in the context of this claim encompasses the user manually receiving generic “input data” and performing generic matrix “generation” steps. If a claim limitation, under its
broadest reasonable interpretation, covers performance of the limitation in the mind but for the
recitation of generic computer components, then it falls within the “Mental Processes” grouping
of abstract ideas (concepts performed in the human mind (including an observation, evaluation,
judgment, opinion)).
Further, these concepts also recite “Certain Methods of Organizing Human Activity”; (such as
commercial or legal interactions (including agreements in the form of contracts; legal
obligations; advertising, marketing or sales activities or behaviors; business relations) where
performing generic determining steps of generic clusters using generic input data is a method of human activity in commercial or legal interactions.
Accordingly, the claim recites an abstract idea.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites one additional element – using a processor with a memory to perform both the acquiring; generating and providing; and determining steps. The processor with a memory in both steps is recited at a high level of generality (i.e., as a generic processor performing a generic computer function of “determining”) such that it amounts no more than mere instructions to apply the exception using a generic computer component.
Accordingly, this additional element does not integrate the abstract idea into a practical
application because it does not impose any meaningful limits on practicing the abstract idea.
The claim is directed to an abstract idea. The claim does not include additional elements that
are sufficient to amount to significantly more than the judicial exception.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of a processor with a memory to perform
both the acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim(s) is/are not patent eligible.
Referring to claim 18, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “the input data includes graph data including a plurality of nodes and edges; and
the edges represent connection relationships for at least some of the plurality of nodes.”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “the input data includes graph data including a plurality of nodes and edges; and the edges represent connection relationships for at least some of the plurality of nodes” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “the input data includes graph data including a plurality of nodes and edges; and the edges represent connection relationships for at least some of the plurality of nodes” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 19, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 20, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the graph data includes graph adjacency information, node feature information, and edge feature information”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the graph data includes graph adjacency
information, node feature information, and edge feature information” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the graph data includes graph adjacency information, node feature information, and edge feature information” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 21, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein determining the number of clusters includes: generating a grouping matrix representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster; determining a pooling operator based on the grouping matrix; and determining the number of clusters based on the pooling operator”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein determining the number of clusters includes: generating a grouping matrix representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster; determining a pooling operator based on the grouping matrix; and determining the number of clusters based on the pooling operator” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein determining the number of clusters includes: generating a grouping matrix representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster; determining a pooling operator based on the grouping matrix; and determining the number of clusters based on the pooling operator” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 22, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the processor is configured to combine the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input graph data, regardless of the numbering order of the collected input graph data”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the processor is configured to combine the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input graph data, regardless of the numbering order of the collected input graph data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the processor is configured to combine the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input graph data, regardless of the numbering order of the collected input graph data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 23, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the generating the grouping matrix includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the generating the grouping matrix includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the generating the grouping matrix includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 24, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the determining the pooling operator includes decomposing the grouping matrix into a square-root form”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the determining the pooling operator
includes decomposing the grouping matrix into a square-root form” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the determining the pooling operator includes decomposing the grouping matrix into a square-root form” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 25, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the processor is configured to generate the pooling operator comprising a number of groups after pooling and the nodes assigned within the same group”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the processor is configured to generate the pooling operator comprising a number of groups after pooling and the nodes assigned within the same group” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the processor is configured to generate the pooling operator comprising a number of groups after pooling and the nodes assigned within the same group” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 26, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein: the pooling operator is in the form of a transformation matrix comprising a status of nodes for each group; and the status of nodes indicates whether the nodes are exactly pooled into the same clusters or pooled orthogonally into different clusters”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein: the pooling operator is in the form of a transformation matrix comprising a status of nodes for each group; and the status of nodes indicates whether the nodes are exactly pooled into the same clusters or pooled orthogonally into different clusters” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein: the pooling operator is in the form of a transformation matrix comprising a status of nodes for each group; and the status of nodes indicates whether the nodes are exactly pooled into the same clusters or pooled orthogonally into different clusters” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Claim 27 recites: (Step 2a, Prong One)
determining a number of clusters for performing pooling based on the input data.
The limitation of determining a number of clusters for performing pooling based on the input data, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, other than reciting a generic method/model, nothing in the claim element precludes the step from practically being performed in the mind. For example, but for the method/model language, “determining” in the context of this claim encompasses the user manually determining a number of generic “clusters” using generic “input data” steps. Similarly, the limitation(s) of
acquiring; generating and providing, as drafted, is a process that, under its broadest reasonable
interpretation, covers performance of the limitation in the mind but for the recitation of generic
computer components. For example, but for the method/model language, acquiring; generating and providing in the context of this claim encompasses the user manually receiving generic “input data” and performing generic matrix “generation” steps. If a claim limitation, under its
broadest reasonable interpretation, covers performance of the limitation in the mind but for the
recitation of generic computer components, then it falls within the “Mental Processes” grouping
of abstract ideas (concepts performed in the human mind (including an observation, evaluation,
judgment, opinion)).
Further, these concepts also recite “Certain Methods of Organizing Human Activity”; (such as
commercial or legal interactions (including agreements in the form of contracts; legal
obligations; advertising, marketing or sales activities or behaviors; business relations) where
performing generic determining steps of generic clusters using generic input data is a method of human activity in commercial or legal interactions.
Accordingly, the claim recites an abstract idea.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites one additional element – using a method/model to perform both the acquiring; generating and providing; and determining steps. The method/model in both steps is recited at a high level of generality (i.e., as a generic method/model performing a generic computer function of “determining”) such that it amounts no more than mere instructions to apply the exception using a generic computer component.
Accordingly, this additional element does not integrate the abstract idea into a practical
application because it does not impose any meaningful limits on practicing the abstract idea.
The claim is directed to an abstract idea. The claim does not include additional elements that
are sufficient to amount to significantly more than the judicial exception.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of a method/model to perform
both the acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim(s) is/are not patent eligible.
Additionally, Claims 27-36 are rejected under 35 U.S.C. 101 because the claimed invention is
Directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the
four categories of patent eligible subject matter because independent claim(s) 27 does not recite statutory computer hardware/processors (only generic methods/”method for learning an artificial intelligence model” without even generically described computer processing elements, for example a “computer-implemented model”) without limitation and thus the claim(s) is/are directed to a signal per se and/or mere information in the form of data, and dependent claims 28-36 do not correct this deficiency.
See generally guidance on the New Form Paragraphs for Subject Matter Eligibility Rejections
under the 2019 Revised Patent Subject Matter Eligibility Guidance (¶ 7.05.01 Rejection, 35
U.S.C. 101, Nonstatutory (Not One of the Four Statutory Categories);
Available via:
https://www.uspto.gov/sites/default/files/documents/form_para_for_2019peg_20190108.pdf
Referring to claim 28, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein: the input data includes graph data including a plurality of nodes and edges; and the edges represent connection relationships for at least some of the plurality of nodes”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein: the input data includes graph data including a plurality of nodes and edges; and the edges represent connection relationships for at least some of the plurality of nodes” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein: the input data includes graph data including a plurality of nodes and edges; and the edges represent connection relationships for at least some of the plurality of nodes” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 29, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 30, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the graph data includes graph adjacency information, node feature information, and edge feature information”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the graph data includes graph adjacency
information, node feature information, and edge feature information” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the graph data includes graph adjacency information, node feature information, and edge feature information” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 31, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein determining the number of clusters includes: generating a grouping matrix representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster; determining a pooling operator based on the grouping matrix; and determining the number of clusters based on the pooling operator”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein determining the number of clusters includes: generating a grouping matrix representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster; determining a pooling operator based on the grouping matrix; and determining the number of clusters based on the pooling operator” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein determining the number of clusters includes: generating a grouping matrix representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster; determining a pooling operator based on the grouping matrix; and determining the number of clusters based on the pooling operator” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 32, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “further comprising combining the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input data, regardless of the numbering order of the collected input data”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “further comprising combining the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input data, regardless of the numbering order of the collected input data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “further comprising combining the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input data, regardless of the numbering order of the collected input data” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 33, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the generating the grouping matrix includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the generating the grouping matrix includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the generating the grouping matrix includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 34, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the determining the pooling operator includes decomposing the grouping matrix into a square-root form”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the determining the pooling operator
includes decomposing the grouping matrix into a square-root form” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the determining the pooling operator includes decomposing the grouping matrix into a square-root form” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 35, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein the determining the pooling operator comprises generating the pooling operator comprising a number of groups after pooling and the nodes assigned within the same group”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein the determining the pooling operator
comprises generating the pooling operator comprising a number of groups after pooling and the
nodes assigned within the same group” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein the determining the pooling operator comprises generating the pooling operator comprising a number of groups after pooling and the nodes assigned within the same group” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Referring to claim 36, (Step 2a, Prong One) this further merely performs an additional abstract
mental step of “wherein: the pooling operator is in the form of a transformation matrix comprising a status of nodes for each group; and the status of nodes indicates whether the nodes are exactly pooled into the same clusters or pooled orthogonally into different clusters”.
(Step 2a, Prong Two)
This judicial exception is not integrated into a practical application. In particular, the claim only
recites the additional elements of “wherein: the pooling operator is in the form of a transformation matrix comprising a status of nodes for each group; and the status of nodes indicates whether the nodes are exactly pooled into the same clusters or pooled orthogonally into different clusters” steps to perform both the aforementioned acquiring; generating and providing; and determining steps. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
(Step 2b)
The claim does not include additional elements that are sufficient to amount to significantly
more than the judicial exception. As discussed above with respect to integration of the abstract
idea into a practical application, the additional element of using “wherein: the pooling operator is in the form of a transformation matrix comprising a status of nodes for each group; and the status of nodes indicates whether the nodes are exactly pooled into the same clusters or pooled orthogonally into different clusters” steps to perform both the aforementioned acquiring; generating and providing; and determining steps amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept.
The claim(s) is/are not patent eligible.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 17-18, 20-21, 25 and 27-28, 30-31, 35 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al., US Pub. No. 2022/0101140 A1, in view of Shang et al., US Pub. No. 2021/0383205 A1.
As to claim 17 (and substantially similar claim 27), Kumar discloses:
a system comprising: a memory; and a processor in communication with the memory,
(Kumar abstract and [0018,0084])
wherein the processor is configured to:
acquire input data;
(Kumar [0032] Extracting block 102 may be configured to extract features from input data, such as training data.; see also [0036] The max-pooled features at each level may be used, for
example, to derive a generalized set of feature vectors describing the input data (e.g. text, images, or other values). See also [0038] FIG. 2 illustrates a block diagram showing an
exemplary convolutional and max pooling blocks of a CNN model comprising the feature extractor of the CNN model. As shown, input data 202 (in matrix form) may be passed to
a first layer 204 of ( e.g. convolutional) filters in the CNN model, which is then passed to a first max pooling layer 206.)
automatically determine, based on the input data, a number of clusters for performing
pooling;
(Kumar teaches using clustering to cluster the features, which have been extracted from the max pooling layer outputs see [0055] This dominance determining procedure, as just
described, is illustrated by FIGS. 3 and 4. For example, FIG. 3 shows using clustering to cluster the features, which have been extracted from the max pooling layer outputs; forming
the clustering matrix;
see also [0014] According to a first aspect, a method for explaining deep-learning models is provided. The method includes extracting a set of features from a first deep-learning model
for a first set of training data; clustering the set of features into N groups, wherein N represents a number of unique labels in the first set of training data; forming a clustering matrix from the N groups; and determining dominant columns in the clustering matrix to form a subset of the set of
features.;
see also [0043] Clustering the feature vectors into N groups where N is number of unique labels in the dataset can help to provide additional information about the model. For example, if there are no clusters, each input needs to be analyzed to understand the feature in the input. This is
computationally complex. Therefore, by grouping the feature vectors into clusters, the computational complexity can be reduced.; and [0050] Consider the following procedure for learning the important features. First, construct a matrix with all max pooling layer outputs for each of the input data (this can be referred to as the "max pooling matrix" or alternatively the
"clustering matrix").)
generate a first pooling matrix based on the number of clusters;
(Kumar [0050] Consider the following procedure for learning the important features. First, construct a matrix with all max pooling layer outputs for each of the input data (this can be
referred to as the "max pooling matrix" or alternatively the "clustering matrix"). For discussion purposes, assume the matrix is of size MxN, i.e. there are N elements coming from
max pooling layer outputs, coming from M data points.)
Kumar does not disclose:
and
provide the first pooling matrix as input to an artificial intelligence model to output a
prediction result relating to the input data;
However, Shang discloses:
and
provide the first pooling matrix as input to an artificial intelligence model to output a
prediction result relating to the input data
(Shang teaches a taxonomy manager and a ML manager for using a pooling layer generates an adjacency matrix, Ac, which then is used with a GNN operation, for predicting the is-a relationships between term pairs
i.e. “provide the first pooling matrix as input to an artificial intelligence model to output a
prediction result relating to the input data”
(see 0047-0049)
See [0047]
PNG
media_image2.png
212
354
media_image2.png
Greyscale
See also [0049] While predicting the is-a relationships between term pairs, the ML model (156 A) ensures that the formed resulting graph does not contain cycles nor is disconnected, which in an exemplary embodiment is ensured by adding constraint based regularizers. Accordingly, the taxonomy manager (158) takes in the list of terms from the target domain and feeds the
list of terms into the ML model (156 A), and selects only those pairs whose model based probability exceeds a threshold.
See also [0005] A ML model is subject to training to build a taxonomy structure for a target domain. The ML model training uses a graph neural network (GNN) to aggregate neighbors of one or more of the represented nodes into clusters. In addition, the ML model training applies semantic clustering aggregation to the clusters and produces a preliminary taxonomy for
the target domain.;
see also [0030] The tools, including the AI platform (150), or in an embodiment, the tools embedded therein including the NLP manager (152), the graph manager (154), the ML manager
(156), and the taxonomy manager (158) may be configured to receive input from various sources, including but not limited to input from the network (105), and an operatively
coupled knowledge base (160).)
It would have been obvious to one having ordinary skill in the art at the time the time of the
effective filing date to apply a pooling layer generates an adjacency matrix as taught by Shang, to the system of Kumar, since it was known in the art that machine learning systems can provide a taxonomy manager which is operatively coupled to the ML manager where the taxonomy manager produces a preliminary taxonomy through application of the training ML model, e.g. model A, to the cross-domain graph where application of the ML model accumulates and encodes cluster relationships, and predicts connection between the term pairs where the preliminary taxonomy is subject to enrichment. (Shang [0033]).
As to claim 18, Shang as modified discloses the system of claim 17, wherein:
the input data includes graph data including a plurality of nodes and edges; and
the edges represent connection relationships for at least some of the plurality of nodes
(Shang [0041] API1 (222) provides support to construct a cross-domain directed graph based on the extracted term pairs. The constructed graph includes vertices and edges, with the vertices representing terms, and the edges representing relationships between terms. The edges in the constructed graph include both intra-domain and inter-domain edges.;
See also [0005] term pairs are extracted from a corpus, with the extracted term pairs including at least one cross-domain term pair. A cross-domain directed acyclic graph (DAG) is constructed based on the extracted term pairs. The graph includes vertices and intra-domain and interdomain edges.).
As to claim 20, Shang as modified discloses the system of claim 18, wherein the graph data includes graph adjacency information,
(Shang [0046] A pooling layer uses the adjacency matrix, A, and the node feature matrix, H1,
to generate the soft cluster assignment matrix, S1)
node feature information, and
(Shang Fig. 5 item 502: “soft cluster assignment matrix, S , is generated
at layer, .using the GNN strategy over the adjacency matrix, A, and the node feature matrix,”)
(Shang [0046] A pooling layer uses the adjacency matrix, A, and the node feature matrix, H1,
to generate the soft cluster assignment matrix, S1)
edge feature information.
(Shang Fig. 6 item 602: “Concatenate term embeddings and edge features to generate a vector v . pair”; see also [0050] Term embeddings and edge features, also known as relation
features, are concatenated (602). The variable, v, represents a term, and v hypo and v hyper are the term embeddings for a candidate hyponym-hypernym pair.).
As to claim 21, Shang as modified discloses the system of claim 18, wherein determining the number of clusters includes:
generating a grouping matrix
(Shang teaches generating/aggregating a node representation matrix/ soft cluster assignment
matrix, from an initial/adjacency matrix, i.e. generate “a grouping matrix”
See [0044] An initial matrix, H0 , is randomly initialized from a standard normal distribution (
408). The adjacency matrix, A, and the node representation matrix, H1, are used to iteratively
update the representation of a particular node by aggregating representations of its neighbors
using a traditional GNN strategy (410). A graph neural network (GNN) is used for the iterative
update;
See also [0046] A pooling layer uses the adjacency matrix, A, and the node feature matrix, H1,
to generate the soft cluster assignment matrix, S1
See also [0047] Using the soft cluster assignment matrix, S1 , the pooling layer generates an
adjacency matrix, Ac, for the cluster graph (504) and generates cluster node representations,
H/, for the cluster graph;
See also [0031] The model training further includes application of a semantic clustering
aggregation to the clusters. The semantic cluster aggregation includes application of cluster
based pooling and unpooling to generate node representations that possess latent cluster
information. The cluster based pooling creates a cluster graph that comprises a set of cluster
nodes with representations learned based on a trainable cluster assignment matrix.)
representing, for each pair of nodes in the input data, a probability that the pair belongs to a same cluster;
(Shang [0013] FIG. 6 depicts a flow chart to illustrate a process for estimating the edge probability with respect to representation of a valid hypernym relationship.)
determining a pooling operator based on the grouping matrix;
(Shang [0023] At a basic level, each layer of the neural network includes one or more operators or functions operatively coupled to output and input.; see also [0047] Using the soft cluster assignment matrix, S1 , the pooling layer generates an adjacency matrix, Ac, for the
cluster graph (504) and generates cluster node representations,)
and
determining the number of clusters based on the pooling operator
(Shang [0047] Using the soft cluster assignment matrix, S1, the pooling layer generates an adjacency matrix, Ac, for the cluster graph (504) and generates cluster node representations,
H/, for the cluster graph; see also [0045-0046] [0045] The semantic clustering aggregation at step (310) is the second of two strategies used for embedding generation. Semantic clustering aggregation at step (310), which operates on output from the neighborhood aggregation at step
(308), utilizes soft clustering based pooling and unpooling that uses semantic clustering aggregation for learning model representations. [0046] Analogous to an auto-encoder, the pooling adaptively creates a cluster graph comprising a set of cluster nodes whose representations are learned based on a trainable cluster assignment matrix).
As to claim 25, Shang as modified discloses the system of claim 23, wherein the processor is configured to generate the pooling operator comprising a number of groups after pooling and the nodes assigned within the same group.
(Shang teaches soft cluster assignment matrix, S1, is calculated based on node embeddings,
nodes with similar features and nc is the number of clusters, i.e. “pooling operator comprising a
number of groups after pooling and the nodes assigned” see para. [0046] Analogous to an autoencoder,
the pooling adaptively creates a cluster graph comprising a set of cluster nodes whose
representations are learned based on a trainable cluster assignment matrix. Mathematically, a
soft cluster assignment matrix, S1ER nxn", is generated at layer 1 using the GNN strategy over
the adjacency matrix, A, and the node feature matrix, H1 (502), wherein nc is the number of
clusters. Each row in S1 corresponds to one of n nodes in layer 1, and each column correspond
to one of the nc clusters. A pooling layer uses the adjacency matrix, A, and the node feature
matrix, H1, to generate the soft cluster assignment matrix, S1 , as:
S1~softmax(GNN1,c1u,,er(A,H1 )) where the softmax is a row-wise softmax function, and
ecluste/ER dzxnr denotes all trainable parameters in GNNI cluster. Since the soft cluster
assignment matrix, S1, is calculated based on node embeddings, nodes with similar features
and local structure will have a similar cluster assignment.).
Referring to claim 28, this dependent claim recites similar limitations as claim 18;
therefore, the arguments above regarding claim 18 are also applicable to claim 28.
Referring to claim 30, this dependent claim recites similar limitations as claim 20;
therefore, the arguments above regarding claim 20 are also applicable to claim 30.
Referring to claim 31, this dependent claim recites similar limitations as claim 21;
therefore, the arguments above regarding claim 21 are also applicable to claim 31.
Referring to claim 35, this dependent claim recites similar limitations as claim 25;
therefore, the arguments above regarding claim 25 are also applicable to claim 35.
Claim(s) 19 and 29 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al., US Pub. No. 2022/0101140 A1, in view of Shang et al., US Pub. No. 2021/0383205 A1, in view of Pisner et al, US Pub. No.: US 2020/0167694.
As to claim 19,
Kumar/Shang do not disclose:
wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data;
However, Pisner discloses:
the system of claim 18, wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data, and authorship data
(Pisner teaches hierarchical multigraphs for molecular neural network and social network, i.e. “wherein the graph data has a hierarchical structure and includes at least one of a molecular graph, social network data” see [0044] FIG. 1 is a visualization of a hierarchical multigraph spanning seven independent data modalities for which the disclosed invention is equipped to accommodate (from bottom to top: genetic transcriptome/gene regulatory network, molecular neural network, microstructural brain network, functional brain network, cognitive/semantic network, behavioral network, social network).
See also [0018] A semantic network is used when one has knowledge that is best understood as a set of concepts that are related to one another. This knowledge representation, residing
within the cognitive domain of individual persons, consist of arcs and nodes which can be organized into a taxonomic hierarchy. Semantic networks can be directed or undirected graphs consisting of vertices, which represent concepts, and edges, which represent semantic relations
between concepts, mapping or connecting semantic fields. In practice, semantic networks can be computationally generated using various forms of Natural Language Processing
(NLP) which can parse metadata from text information acquired from audio recordings, text messages, emails, and other sources of semantic information )
It would have been obvious to one having ordinary skill in the art at the time the time of the
effective filing date to apply hierarchical multigraphs for molecular neural network and social network as taught by Pisner, to the system of Kumar/Shang, since it was known in the art that machine learning systems can provide Hierarchical network analysis, supported by the
stabilizing effect of ensemble sampling, which provides a potentially invaluable method for embedding connectomes into machine-learning models where novel connectome-generating tools are need to produce network-enriched features of multiple variations, resolutions,
and scales that can collectively be incorporated into the statistical learning algorithms available through Torch, H20, Scikit-Learn, TensorFlow, or other emerging machine learning
libraries. (Pisner [0022]).
Referring to claim 29, this dependent claim recites similar limitations as claim 19;
therefore, the arguments above regarding claim 19 are also applicable to claim 29.
Claim(s) 22, 32 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al., US Pub. No. 2022/0101140 A1, in view of Shang et al., US Pub. No. 2021/0383205 A1, in view of Gal et al., US Pub. No. 2021/0248367.
As to claim 22, Kumar/Shang do not disclose:
wherein the processor is configured to combine the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input graph data, regardless of the numbering order of the collected input graph data;
However, Gal discloses the system of claim 21, wherein the processor is configured to combine the grouping matrix and the first pooling matrix to obtain, through a Feed-Forward Network, an output graph in which one group operates as one node, the output graph reflecting the hierarchical structure of the collected input graph data, regardless of the numbering order of the collected input graph data
(Gal teaches Graph Convolutional Networks (a feed forward network) for graph representation of documents see [0001] the disclosure relates to Graph Convolutional Networks (GCN) for processing pseudo-spatial graph representations of the underlying structure of documents.;
See also [0050] The resulting n1+1 'feature' entries for each node constitute a soft mapping from the node at layer I to the n1+1 new nodes 304 at layer 1+1. This mapping is used to converge features from multiple nodes into a single node, or to spread the features from one node amongst several new nodes.
See also [0036] The invention propose a new extension to the GCN framework, named Cardinal Graph Convolutional Networks) CGCNs), wherein the graph representation of the
document is built to retain knowledge of cardinal-direction relationships between nodes ('north-of', 'west-of' , ... ) and the network model itself is built to exploit such information. Furthermore, the CGCN framework is built to exploit graph pooling methods, creating a fully convolutional-deconvolutional model and extending the receptive field of the network's filters.;
see also [0008] Graph convolutional networks (GCNs). GCNs (Kipf and Welling, 2016) are an incarnation of graph neural networks (GNNs) that use convolutions (linear transformations) to hierarchically learn node features from one layer to another in a feed-forward neural network).
It would have been obvious to one having ordinary skill in the art at the time the time of the
effective filing date to apply Graph Convolutional Networks for graph representation of documents as taught byGal, to the system of Kumar/Shang, since it was known in the art that machine learning systems can provide image-based document analysis using Graph Convolutional Networks (GCN) for processing pseudo-spatial graph representations of the underlying structure of documents where Cardinal Graph Convolutional Networks (CGCN)
are an efficient and flexible extension of GCNs with cardinal- direction awareness of spatial node arrangement, where before no such capability existed where the new mathematical formulation of CGCNs retains the traditional GCN permutation invariance, ensuring directional neighbors are intrinsically involved in learning abstract representations, even in the absence of a proper ordering of the nodes where CGCNs achieve state of the art results on an invoice information extraction task, jointly learning a word level tagging as well as document meta-level regression problem. (Gal [0031-0032])
Referring to claim 32, this dependent claim recites similar limitations as claim 22;
therefore, the arguments above regarding claim 22 are also applicable to claim 32.
Claim(s) 23, 33 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al., US Pub. No. 2022/0101140 A1, in view of Shang et al., US Pub. No. 2021/0383205 A1, in view of Taslakian et al., US Pub. No.: US 2023/0025826.
As to claim 23, Kumar/Shang do not disclose:
wherein the generating the grouping matrix
includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1;
However, Taslakian discloses
the system of claim 21, wherein the generating the grouping matrix
includes grouping the plurality of nodes and the edges based on similarity according to a preset grouping criterion, and designating identifiers ranging from 0 to 1.
(Taslakian teaches representing connections between nodes represent paths as scaled output
values with associated weights (e.g., values between 0 and 1, inclusive, i.e. “connection
relationship as identifiers of 0 or more to 1 or less based on similarity” see para. [0147]
Connections between nodes represent paths through which intermediate values flow, and are
each associated with a respective weight that is applied to the respective intermediate value.
Each node performs an operation on its received values and their associated weights (e.g.,
values between 0 and 1, inclusive) to produce an output value. In some cases this operation
may involve a dot-product sum of the products of each input value and associated weight. An
activation function ( e.g., a sigmoid, tan h or ReLU function) may be applied to the result of the
dot-product sum to produce a scaled output value.;
see also [0240] In some embodiments, the embedding of the specific node is based on
multiplication of the attributes of the specific node by a first matrix of weights and multiplication
of the attributes of the one or more neighboring nodes by a second matrix of weights.).
It would have been obvious to one having ordinary skill in the art at the time the time of the
effective filing date to applying scaled output values representing connections between nodes
represent paths with associated weights (e.g., values between 0 and 1) as taught by Taslakian
to the system of Shang / Bianchi / Song since it was known in the art that machine learning
systems can provide producing a scaled output value for determining error values are
determined for all of the sets of input values, and the error function involves calculating an
aggregate ( e.g., an average) of these values where once the error is determined, the weights
on the connections are updated in an attempt to reduce the error where this update process
should reward "good" weights and penalize "bad" weights and thus, the updating should
distribute the "blame" for the error through the neural network in a fashion that results in a lower
error for future iterations of the training data. (Taslakian [0147-0149]).
Referring to claim 33, this dependent claim recites similar limitations as claim 23;
therefore, the arguments above regarding claim 23 are also applicable to claim 33.
Claim(s) 24, 26 and 34 and 36 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kumar et al., US Pub. No. 2022/0101140 A1, in view of Shang et al., US Pub. No. 2021/0383205 A1, in view of Taslakian et al., US Pub. No.: US 2023/0025826, in view of Song et al., Article: “Why Approximate Matrix Square Root Outperforms Accurate
SVD in Global Covariance Pooling”; Proceedings of the IEEE/CVF International
Conference on Computer Vision (ICCV), 2021, pp. 1115-1123; openaccess.thecvf.com.
As to claim 24, Shang/Kumar/ Taslakian do not disclose:
wherein the determining the pooling operator includes decomposing the grouping matrix into a square-root form;
However, Song discloses the system of claim 21, wherein the determining the pooling operator includes decomposing the grouping matrix into a square-root form
(Song teaches Global Covariance Pooling (GCP) using SVD
decomposition/eigendecomposition and using the matrix square root, i.e. “the grouping matrix
is decomposed in a square-root form to obtain a pooling operator” see p. 1115 sec 1.
Introduction “1. Introduction Global Covariance Pooling (GCP) explores the second order
statistics by normalizing the covariance matrix of the convolutional features before feeding them
to the fully connected layer. It has been shown to outperform the
first-order pooling methods (e.g., max-pooling and averagepooling) [9, 17, 15, 16, 14].
Generally, a GCP metalayer computes the covariance matrix of the features as the global
representation, and then performs eigendecomposition to derive the corresponding eigenvalues
and eigenvectors, followed by normalization using either matrix logarithm [9, 17] or the matrix
square root”;
See also p. 1116 “This has enlightened us to combine the SVD function with iSQRTCOV
[14] and to develop a hybrid training protocol, i.e., use Newton-Schulz iteration to train the
network until the learning rate is sufficiently small and the network weights are relatively stable,
then switch to the ordinary/modified SVD for accurate matrix square root calculation. By doing
so, SVD only deals with well-conditioned matrices in the later stage. This hybrid strategy fully
explores the potential of SVD for eigendecomposition, and thus these SVD methods achieve
competitive and sometimes better performance than iSQRT-COV”).
It would have been obvious to one having ordinary skill in the art at the time the time of the
effective filing date to applying Global Covariance Pooling (GCP) using SVD decomposition and
using the matrix square root as taught by Song to the system of Shang / Kumar / Taslakian since it was known in the art that machine learning systems can provide an approximate matrix square root which outperforms exact ones on GCP from data precision and gradient smoothness where various SVD remedies with smooth gradients and validate their performances on GCP and a hybrid training strategy helps the SVD methods to achieve competitive performances against Newton-Schulz iteration and where based on the findings, a new GCP meta-layer that uses SVD as the forward pass and Pade approximants during back-propagation for robust gradient approximation where a proposed meta-layer has achieved state-of-the-art performances on different datasets and deep models. (Song sec. 6. Conclusion).
As to claim 26, Shang as modified discloses the system of claim 25, wherein:
the pooling operator is in the form of a transformation matrix comprising a status of
nodes for each group;
(Shang teaches node labels/classification/features and a cluster
assignment matrix, i.e. “a transformation matrix comprising a status of nodes for each group”
see [0028] A graph neural network (GNN) is a type of neural network which directly operates on
the graph structure. More specifically, the GNN functions on input defined in the
language of nodes, edges, and neighborhoods, and utilizes message passing between nodes.
Message passing includes collecting information from neighbors, aggregating the collected
information, and update self-state and weights based on the aggregated representation. A
typical application of the GNN is a node classification. Essentially, every node, v, in the graph is
characterized by its feature, x, and associated with a label, t. There are two primary processes
for the taxonomy construction, including a training cycle and an inference cycle.;
see also [0032] The semantic cluster aggregation includes application of cluster based pooling
and unpooling to generate node representations that possess latent cluster information. The
cluster based pooling creates a cluster graph that comprises a set of cluster nodes with
representations learned based on a trainable cluster assignment matrix. The cluster based
unpooling decodes the created cluster graph into an original graph using the learned cluster
assignment matrix.).
and Song, as modified discloses:
the status of nodes indicates whether the nodes are exactly pooled into the same
clusters or pooled orthogonally into different clusters.
(see Song p. 1117 sec. 3.1:
PNG
media_image3.png
638
422
media_image3.png
Greyscale
see also Song p. 1121 sec. 4.4:
PNG
media_image4.png
284
442
media_image4.png
Greyscale
)
Referring to claim 34, this dependent claim recites similar limitations as claim 24;
therefore, the arguments above regarding claim 24 are also applicable to claim 34.
Referring to claim 36, this dependent claim recites similar limitations as claim 26;
therefore, the arguments above regarding claim 26 are also applicable to claim 36.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
CONTACT INFORMATION
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EVAN S ASPINWALL whose telephone number is (571)270-7723. The examiner can normally be reached Monday-Friday 8am-5pm.
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, Neveen Abel-Jalil can be reached at (408)918-7548. 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.
/Evan Aspinwall/Primary Examiner, Art Unit 2156
Prosecution Insights