DETAILED ACTION
Claims 1-23 are presented for examination.
This office action is in response to submission of application on 19-JANUARY-2026.
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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 30-JANUARY-2024 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Amendment
The amendment filed 19-JANUARY-2026 in response to the non-final office action mailed 17-OCTOBER-2025 has been entered. Claims 1-23 remain pending in the application.
With regards to the non-final office action’s rejection under 103, the amendment to the claims have overcome the original rejection. However, upon a new search for the amended limitations, a new 103 rejection over Choi in view of Wu, further in view of new art Potharaju has been written. In light of the new art, the arguments are rendered moot.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 6-12, 14-19, and 21-23 are rejected under 35 U.S.C. 103 as being unpatentable over Choi et al. (Pub. No. US 20190080232 A1, filed September 8th 2017, hereinafter Choi) in view of Wu et al. (Pub. No. WO 2022076234 A1, filed September 30th 2021, hereinafter Wu) further in view of Potharaju et al. (Pub. No. US 10877795 B2, filed July 25th 2018, hereinafter Potharaju).
Regarding claim 1:
Claim 1 recites:
A system for estimating throughput for placement graphs for a reconfigurable dataflow computing system comprises: a training module configured to obtain a set of reference placement graphs for one or more computing tasks, each reference placement graph including nodes corresponding to configurable units of the reconfigurable dataflow computing system and edges corresponding between the configurable units; the training module configured to conduct the one or more computing tasks on the reconfigurable dataflow computing system using each reference placement graph of the set of reference placement graphs to determine a measured throughput value for each reference placement graph of the set of reference placement graphs; the training module configured to train a graph neural network using each corresponding throughput value as a training target to produce a trained graph neural network; an estimation module configured to configure the trained graph neural network for a candidate placement graph corresponding to a target computing task; the estimation module configured to use the trained graph neural network to estimate a throughput for the target computing task conducted on the reconfigurable dataflow computing system according to the candidate placement graph
Regarding the limitation system for estimating throughput for placement graphs for a reconfigurable dataflow computing system:
Choi teaches estimating a given deep neural network’s (DNN) performance on particular hardware architectures using a variety of data flows (Paragraph 21) wherein the DNN may be represented as a graph (Paragraph 83). This would make the DNN’s description that is used as input a placement graph as it describes a particular part of the overall resources available as described in the present application’s specification, paragraph 106. Furthermore, the variety of dataflows demonstrates that the dataflow is reconfigurable as multiple are available.
Choi discloses a training module configured to obtain a set of reference placement graphs for one or more computing tasks, each reference placement graph including nodes corresponding to configurable units of the reconfigurable dataflow computing system and edges corresponding between the configurable units; the training module configured to conduct the one or more computing tasks on the reconfigurable dataflow computing system using each reference placement graph of the set of reference placement graphs to determine [a measured throughput value] for each reference placement graph of the set of reference placement graphs:
Choi teaches that for the various configurations of a deep neural network, or the obtained set of reference placement graphs for one or more computing tasks as each DNN would be analogous to a placement graph as seen above, a metric of interest is calculated for each of the set of deep neural network configuration, i.e. each reference placement graph of the set of reference placement graphs. (Paragraph 55). Furthermore, the metric of interest may clearly be throughput, as the invention of Choi is directed as solving challenges in addressing throughput goals (Paragraph 3), and therefore calculating the metric of interest would be determining a corresponding throughput value for each placement graph. The estimation itself would be one or more computing tasks.
However, Choi does not teach each reference placement graph including nodes corresponding to configurable units of the reconfigurable dataflow computing system and edges corresponding between the configurable units which is taught by Potharaju below.
Choi is supplemented by Potharaju below in order to explicitly teach a throughput.
Choi discloses an estimation module configured to configure the trained graph neural network for a candidate placement graph corresponding to a target computing task; the estimation module configured to use the trained graph neural network to estimate [a throughput for the target computing task] conducted on the reconfigurable dataflow computing system according to the candidate placement graph:
Choi teaches a system that may evaluate a metric of interest, which may be a throughput, for a possible DNN configuration, which would be a candidate placement graph corresponding to a target computing task (Paragraph 4) as the DNN may be used for a particular purpose which would be the target computing task.
Furthermore, this limitation in Choi would also estimate the metric of interest for the target computing task conducted on the reconfigurable dataflow computing system (i.e., the system of Choi as previously discussed), according to the deep neural network candidate configuration, which would be the candidate placement graph.
Choi alone does not teach a trained graph neural network. This aspect of the limitation is taught by Wu, below.
Choi is supplemented by Potharaju below in order to explicitly teach a throughput.
Potharaju in the same field of endeavor of machine learning discloses each reference placement graph including nodes corresponding to configurable units of the reconfigurable dataflow computing system and edges corresponding between the configurable units:
Potharaju teaches a dataflow computational graph wherein operators are represented as nodes (Column 1, lines 30-35), wherein hardware units may be a type of operator i.e. node represented by the graph with connected edges between them (Column 12, lines 43-50), as Potharaju in that section teaches that an “executable component” may be hardware units, wherein an operator may be an executable component. Furthermore, Potharaju describes an embodiment wherein these operators are configurable (Column 15, lines 10-15) as well as teach that the graph itself represents a reconfigurable dataflow computing system (Column 2, lines 1-10).
Potharaju is analogous art to the present application because they are in the same field of endeavor of machine learning.
Potharaju teaches determining a measured throughput value:
Potharaju teaches monitoring a performance parameter of its dataflow execution graph, which would be it reference placement graph. One example of this parameter is a throughput, wherein the monitoring determines a measured throughput value (Column 4, lines 55-67 and Column 5, lines 1-5).
Wu discloses the training module configured to train a graph neural network using each corresponding throughput value as a training target to produce a trained graph neural network:
Wu in the same field of endeavor of machine learning teaches the use of a graph neural network (Paragraph 121) which would include its training.
Furthermore, Choi has previously taught the evaluation of a metric of interest which may be a throughput, wherein the evaluations are used to maximize the metric (Paragraph 4). Therefore, the evaluated throughputs are be used as input and a training target. Therefore, Choi in combination with Wu may produce a system wherein the throughput value of Choi is used with the graph neural network of Wu.
Choi and Wu are analogous art to the present application because they are all in the same field of endeavor of machine learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57) as well as reducing resource underutilization (Potharaju, Column 1, lines 25-30).
Regarding claim 2, which depends upon claim 1:
Claim 2 recites:
The system of claim 1, further comprising a configuration module configured to generate configuration information that enables the reconfigurable dataflow computing system to conduct the target computing task according to the candidate placement graph
Choi in view of Wu, further in view of Potharaju discloses the system of claim 1 upon which claim 2 depends. Furthermore, regarding the limitations of claim 2:
Choi teaches that in order to implement the target computing task according to the candidate placement graph as seen above in claim 1, the DNN is mapped onto the configurable hardware architecture, wherein this is performed by a particular framework that would act as the configuration module (Paragraph 34). This generates configuration information that enable the reconfigurable dataflow computing system to conduct the task, as the hardware would be the computing system that is being reconfigured in order to support the DNN.
Regarding claim 3, which depends upon claim 2:
Claim 3 recites:
The system of claim 2, further comprising a control module configured to configure the reconfigurable dataflow computing system using the configuration information
Choi in view of Wu, further in view of Potharaju discloses the system of claim 2 upon which claim 3 depends. Furthermore, regarding the limitations of claim 3:
Choi teaches that its system may iterate through each dataflow supported by the architecture (Paragraph 76) wherein the determination of which dataflows are supported would be through means of the configuration information and the architecture is the reconfigurable dataflow computing system. Therefore iterating through the dataflows supported by the architecture would involve configuring the architecture for those data flows using the configuration information.
Regarding claim 4, which depends upon claim 3:
Claim 4 recites:
The system of claim 3, wherein the control module is configured to launch execution of the target computing task with the reconfigurable dataflow computing system according to the candidate placement graph
Choi in view of Wu, further in view of Potharaju discloses the system of claim 3 upon which claim 4 depends. Furthermore, regarding the limitations of claim 4:
Choi further teaches that the iteration of the supported dataflows are done in accordance with the design space (Paragraph 76), which is the possible configurations of DNNs to map DNN layers to hardware (Paragraph 44). Therefore, this process would take place according the candidate placement graph of the DNN.
The task would be the optimization of the candidate DNNs, which is performed through the iterations of the design space (Paragraph 76). Therefore, the execution of the target computing task is performed.
Regarding claim 6, which depends upon claim 1:
Claim 6 recites:
The system of claim 1, wherein nodes in a placement graph used for training or estimating correspond to a set of configurable units of the reconfigurable dataflow computing system.
Choi in view of Wu, further in view of Potharaju discloses the system of claim 1 upon which claim 6 depends. Furthermore, Potharaju discloses the limitations of claim 6:
Potharaju teaches that the operators of its graphs i.e. the nodes that are units of the graph are configurable (Column 15, lines 10-15) as well as teach that the graph itself represents a reconfigurable dataflow computing system (Column 2, lines 1-10).
Furthermore, this graph is used to estimate a throughput, which would be a use for estimating (Column 4, lines 55-67 and Column 5, lines 1-5).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of reducing resource underutilization (Potharaju, Column 1, lines 25-30).
Regarding claim 7, which depends upon claim 6:
Claim 7 recites:
The system of claim 6, wherein the set configurable units comprises one or more compute units, one or more memory units and one or more switch units of the reconfigurable dataflow computing system.
Choi in view of Wu, further in view of Potharaju discloses the system of claim 6 upon which claim 7 depends. Furthermore, Choi and Potharaju discloses the limitations of claim 7:
Choi teaches a network that may comprise switches, gateway computers (compute units) and edge servers (which may be memory units as they are used for storage) (Paragraph 126).
Potharaju teaches a reconfigurable dataflow computing system (Column 2, lines 1-10) wherein hardware units are used as operators (Column 12, lines 43-50) Therefore, in combination with Choi, the hardware units described by Choi may be used in the reference placement graph as taught by Potharaju in order to teach compute, memory, and switch units that are represented as nodes in such a graph.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of reducing resource underutilization (Potharaju, Column 1, lines 25-30).
Regarding claim 8, which depends upon claim 6:
Claim 8 recites:
The system of claim 6, wherein an embedding stage of the graph neural network and the trained graph neural comprises a branch for each configurable unit of the set of configurable units
Choi in view of Wu, further in view of Potharaju discloses the system of claim 6 upon which claim 8 depends. Furthermore, regarding the limitations of claim 8:
Choi teaches that when the DNN is represented as a graph, each of the layers may be represented by nodes (Paragraph 83). Therefore, the configurable units of the set of configurable units comprise branches of the graph through their edges as part of the representation of the structure, which would be considered an embedding stage as an embedding reduces something in complexity as the DNN is in graph form.
Choi does not teach a graph neural network, but Wu previously has in claim 1.
Regarding claim 9, which depends upon claim 8:
Claim 9 recites:
The system of claim 8, wherein inputs to each branch of the embedding stage comprise a set of configuration unit attributes for a configuration unit of the set of configurable units
Choi in view of Wu, further in view of Potharaju discloses the system of claim 8 upon which claim 9 depends. Furthermore, regarding the limitations of claim 9:
Choi teaches that the branches of the embedding stage are layers of the DNN (Paragraph 83), and as such would receive inputs that were a set of configuration unit attributes for a configuration unit of the set of configurable units, as the DNN layers would receive weights and inputs from the previous layer on the branch (Paragraph 25) which would be associated with the other layers, or the configuration units of the set of configurable units.
Regarding claim 10, which depends upon claim 9:
Claim 10 recites:
The system of claim 9, wherein the set of configuration unit attributes comprise one or more of configurable unit type, a dataflow task, an end-to-end (e2e) attribute and a routing length
Choi in view of Wu, further in view of Potharaju discloses the system of claim 9 upon which claim 10 depends. Furthermore, regarding the limitations of claim 10:
Choi teaches a variety of parameters for use in its system, including memory parameters which would be a configuration unit attribute that comprises a configurable unit type as it includes the capacity of each memory structure (Paragraph 42). This would be an example of an end-to-end attribute which is defined in the specification as “a sink unit's readiness to receive incoming packets end-to-end from the source. The higher the readiness, the more spaces the unit will have to receive data” (Present application specification, Paragraph 123). The memory capacity measures readiness as it measures the space that the unit has to receive data.
Regarding claim 11, which depends upon claim 9:
Claim 11 recites:
The system of claim 9, wherein each branch of the embedding stage uses a set of embedding tables comprising an embedding table for each configuration unit attribute of the set of configuration unit attributes
Choi in view of Wu, further in view of Potharaju discloses the system of claim 9 upon which claim 11 depends. However, Choi does not teach the limitations of claim 11:
Wu teaches an embedding matrix that includes a feature vector as embeddings, which would be a type of embedding table, for object in a graph (Paragraph 50). The feature vectors for each object would be an embedding table for each configuration unit attribute of the set of configuration unit attributes as features would be analogous to attributes and a feature vector exists for each configuration unit.
Choi previously teaches branches of the embedding stage, where the embedding tables of Wu may be combined with Choi for the reasons listed below.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 12, which depends upon claim 11:
Claim 12 recites:
The system of claim 11, wherein each branch of the embedding stage generates a composite feature vector from a set of embedding vectors provided by the set of embedding tables used by the branch
Choi in view of Wu, further in view of Potharaju discloses the system of claim 11 upon which claim 12 depends. However, Choi does not teach the limitations of claim 12:
Wu teaches the generation of a combined feature vector may be formed by aggregating information of nodes (Paragraph 69) wherein the nodes are associated with particularly reference objects that have their own associated feature vectors, i.e., are from a set of embedding vectors provided by the set of embedding tables (Paragraph 67). Therefore, the combined feature vector of Wu would be analogous to the composite feature vector. This may be considered a branch of the embedding stage as the graph on which these nodes are included may only be a subset of the set of reference objects (Paragraph 66) therefore not including every embedded feature vector.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 14, which depends upon claim 12:
Claim 14 recites:
The system of claim 12, wherein the set of embedding tables comprise one or more of a configurable unit type table, a dataflow task table, an e2e attribute table and a routing length table
Choi in view of Wu, further in view of Potharaju discloses the system of claim 12 upon which claim 14 depends. However, Choi does not teach the limitations of claim 14:
Wu teaches that the feature vector, which is the embedding table as previously discussed in claim 11, may include characteristics of the object (Paragraph 38) which would include the configurable unit type, as the configurable unit type describe the type of object that the unit would be, as would characteristics of the object.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 15, which depends upon claim 8:
Claim 15 recites:
The system of claim 8, wherein the graph neural network and the trained graph neural network comprise a graph aggregation stage
Choi in view of Wu, further in view of Potharaju discloses the system of claim 8 upon which claim 15 depends. However, Choi does not teach the limitations of claim 15:
Wu teaches a graph discriminator (which it considers to be using a graph neural network (Paragraph 78)) wherein aggregation data is obtained from the graph (Paragraph 69). This would be a graph aggregation stage by a graph neural network which is trained.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 16, which depends upon claim 15:
Claim 16 recites:
The system of claim 15, wherein the graph aggregation stage determines an aggregated feature vector for each node in the placement graph to produce aggregated feature vectors
Choi in view of Wu, further in view of Potharaju discloses the system of claim 15 upon which claim 16 depends. However, Choi does not teach the limitations of claim 16:
Wu teaches the generation of a combined feature vector may be formed by aggregating information of nodes around a central (Paragraph 69) wherein the nodes are associated with particularly reference objects that have their own associated feature vectors, i.e., are from a set of embedding vectors provided by the set of embedding tables (Paragraph 67). Therefore, the combined feature vector of Wu would be analogous to the aggregated feature vector for the central node, which may be performed for each node in the graph and its neighbors.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 17, which depends upon claim 16:
Claim 17 recites:
The system of claim 16, wherein the graph aggregation stage determines the aggregated feature vectors by exchanging messages between each pair of connected nodes in the placement graph
Choi in view of Wu, further in view of Potharaju discloses the system of claim 16 upon which claim 17 depends. However, Choi does not teach the limitations of claim 17:
Wu teaches that the information gathered for the aggregation may be gathered from neighboring nodes, wherein the gathering of that information would be a form of messaging between the central node and neighboring node (Paragraph 53). Therefore, the aggregated feature vectors are determined by exchanging message between each pair of connected nodes since the aggregation is based on the edges between the nodes.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 18, which depends upon claim 17:
Claim 18 recites:
The system of claim 17, wherein the graph aggregation stage conducts two or more passes of exchanging messages
Choi in view of Wu, further in view of Potharaju discloses the system of claim 17 upon which claim 18 depends. However, Choi does not teach the limitations of claim 18:
Wu teaches that the messaging as previously describes may also be performed through gathering the information of the k-nearest neighbors, wherein for a k value above 1, there would be two or more passes of exchanging messages in order to determine the nearest neighbors (Paragraph 70).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 19, which depends upon claim 16:
Claim 19 recites:
The system of claim 16, wherein the graph aggregation stage averages the aggregated feature vectors to produce an average feature vector for the placement graph
Choi in view of Wu, further in view of Potharaju discloses the system of claim 16 upon which claim 19 depends. However, Choi does not teach the limitations of claim 19:
Wu teaches that each of the node of the graph have their own respective feature vector, which are aggregated to a central node’s feature vector to form an input feature graph for classification (Paragraph 69). The central’s node aggregated feature vector would be the average feature vector for the placement graph as the central node’s distances from the neighboring nodes determines the weight given to the respective edges connecting it to neighboring nodes, which would produce a weighted average (Paragraph 38).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, Wu, and Potharaju. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57).
Regarding claim 21, which depends upon claim 1:
Claim 21 recites:
The system of claim 1, wherein the training module updates weights within a regressor stage and an embedding stage of the graph neural network via backpropagation.
Choi in view of Wu, further in view of Potharaju discloses the system of claim 1 upon which claim 21 depends. Furthermore, regarding the limitations of claim 21:
Choi teaches that backpropagation is used in order to find the error of the output of the DNN, or the particular throughput estimation (Paragraph 27), which as seen the present specification is how the regressor stage is defined – “During training the estimated (throughput) metric is compared with the measured (throughput) metric and the error is backpropagated to update the weights” (Present application specification, paragraph 128). Backpropagation updates the weights of the neural network inherently.
Claim 22 recites a method that parallels the system of claim 1. Therefore, the analysis discussed above with respect to claim 1 also applies to claim 22. Accordingly, claim 22 is rejected based on substantially the same rationale as set forth above with respect to claim 1.
Claim 23 recites a computer readable medium that parallels the system of claim 1. Therefore, the analysis discussed above with respect to claim 1 also applies to claim 23. Accordingly, claim 23 is rejected based on substantially the same rationale as set forth above with respect to claim 1.
Claims 5, 13, 20 are rejected us 35 U.S.C. 103 as being unpatentable over Choi in view of Wu, further in view of Potharaju, further in view of Li et al. (Pub. No. US 11537719 B2, filed May 17th 2019, hereinafter Li).
Regarding claim 5, which depends upon claim 4:
Claim 5 recites:
The system of claim 4, wherein the configuration information incorporates a selected routing for the candidate placement graph
Choi in view of Wu, further in view of Potharaju discloses the system of claim 4 upon which claim 5 depends. However, neither Choi nor Wu teach the limitations of claim 5:
Li in the same field of endeavor of machine learning teaches determining routing for communicating between elements, wherein the routing may be represented by a graph (Column 19, lines 5-10). This would be an example of selected routing for a candidate placement graph. While Li uses this in a different context, the determination of a routing path for a particular graph is still applicable to Choi in view of Wu, further in view of Potharaju for reasons of the advantage listed below.
Li and the present application are analogous art because they are in the same field of endeavor of machine learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, the teachings of Wu, the teachings of Potharaju, and the teachings of Li. This would have provided the advantage of learning graphs efficiently and improving performance (Li, Column 18, lines 10-15).
Regarding claim 13, which depends upon claim 12:
Claim 13 recites:
The system of claim 12, wherein the feature vector is generated using a multi- layer perceptron
Choi in view of Wu, further in view of Potharaju discloses the system of claim 12 upon which claim 13 depends. However, neither Choi nor Wu teach the limitations of claim 13:
Li teaches that a node encoder neural network, which may be a multi-layer perceptron, may generate a node state representation vector, which would be a kind of feature vector (Column 11, lines 35-45).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, the teachings of Wu, the teachings of Potharaju, and the teachings of Li. This would have provided the advantage of learning graphs efficiently and improving performance (Li, Column 18, lines 10-15).
Regarding claim 20, which depends upon claim 19:
Claim 20 recites:
The system of claim 19, wherein a regressor stage estimates the throughput for the placement graph from the average feature vector using a multi-layer perceptron
Choi in view of Wu, further in view of Potharaju discloses the system of claim 19 upon which claim 20 depends. Furthermore, regarding the limitations of claim 20:
Choi teaches that backpropagation is used in order to find the error of the output of the DNN, or the particular throughput estimation (Paragraph 27), which as seen the present specification is how the regressor stage is defined – “During training the estimated (throughput) metric is compared with the measured (throughput) metric and the error is backpropagated to update the weights” (Present application specification, paragraph 128).
Furthermore, as can be seen in claim 19 upon which this claim depends, Wu teaches the use of a combined feature vector as input, which may be used in combination with the backpropagation of Choi.
Furthermore, Li teaches the use of a multi-layer perceptron (Column 11, lines 35-45) for this limitation to take place on.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to implement a system which utilized the teachings of Choi, the teachings of Wu, the teachings of Potharaju, and the teachings of Li. This would have provided the advantage of improving system efficiency when adapting to new outputs (Wu, Paragraph 57) as well as learning graphs efficiently and improving performance (Li, Column 18, lines 10-15).
Response to Arguments
Applicant’s arguments filed 19-JANUARY-2026 have been fully considered, but the examiner believes that not all are fully persuasive.
Regarding the applicant’s remarks on the non-final office action’s 103 rejection of the claims, the applicant argues that Choi in view of Wu further in view of Li do not teach the amended limitations of these claims. As such, the applicant argues that all claims dependent on the above would additionally not be obvious under 103. The examiner agrees that the prior art of the original office action does not teach the amended limitations. However, upon a new search of the prior art for the amended limitations, the examiner has written a new rejection under 103 to address these limitations and respectfully requests applicant’s consideration of the following:
Regarding the applicant’s argument that “Choi fails to teach “obtaining a set of reference placement graphs” as claimed, and also fails to teach “conduct[ing] the one or more computing tasks … using each reference placement graph … to determine a measured throughput value”. The examiner addresses these limitations through a combination with new art Potharaju:
Potharaju teaches a dataflow computational graph wherein operators are represented as nodes (Column 1, lines 30-35), wherein hardware units may be a type of operator i.e. node represented by the graph with connected edges between them (Column 12, lines 43-50), as Potharaju in that section teaches that an “executable component” may be hardware units, wherein an operator may be an executable component. Furthermore, Potharaju describes an embodiment wherein these operators are configurable (Column 15, lines 10-15) as well as teach that the graph itself represents a reconfigurable dataflow computing system (Column 2, lines 1-10). This therefore describes the reference placement graphs as described above.
Furthermore, Potharaju teaches monitoring a performance parameter of its dataflow execution graph, which would be it reference placement graph. One example of this parameter is a throughput, wherein the monitoring determines a measured throughput value (Column 4, lines 55-67 and Column 5, lines 1-5).
Seeing as Potharaju teaches the recited reference placement graph as detailed above, Choi may be combined with Potharaju in order to incorporate the amended reference placement graphs into the system of Choi.
Potharaju is analogous art to the present application because they are in the same field of endeavor of machine learning. This would have provided the advantage of reducing resource underutilization (Potharaju, Column 1, lines 25-30).
Furthermore, the applicant argues that “the cited portions of Choi do not teach the claimed compute/memory/switch units of such a system that are represented as nodes in a placement graph used for training/estimating throughput”. The examiner addresses these limitations through a combination with new art Potharaju:
As addressed above, Potharaju teaches hardware units that are represented as nodes in a placement graph used for training/estimating throughput (Column 12, lines 43-50) and (Column 4, lines 55-67 and Column 5, lines 1-5). Therefore, in combination with Choi, the hardware units described by Choi may be used in the reference placement graph as taught by Potharaju in order to teach compute, memory, and switch units that are represented as nodes in such a graph.
The applicant argues that “the asserted rationale for combining Choi and Wu (‘improving system efficiency when adapting to new outputs’) is generic and not tied to the specific claim requirements”. However, the examiner believes that this improvement is directly tied towards Wu’s use of a graph neural network, which when incorporated into the system of Choi would enable the described improvement.
The examiner therefore believes that the new art has addressed the deficiencies of the previously presented art with regards to the amendments, and that the rejections of the dependent claims are likewise supported through this reasoning.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALEXANDRIA JOSEPHINE MILLER whose telephone number is (703)756-5684. The examiner can normally be reached Monday-Thursday: 7:30 - 5:00 pm, every other Friday 7:30 - 4:00.
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, Mariela Reyes can be reached at (571) 270-1006. 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.
/A.J.M./Examiner, Art Unit 2142
/Mariela Reyes/Supervisory Patent Examiner, Art Unit 2142
Prosecution Insights