DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of Claims
This Office Action is in response to the communication filed on 30 Jan 2026.
Claims 1-20 are being considered on the merits.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1 and 10 recite the limitation "the encoded distribution" in the third to last limitation. There is insufficient antecedent basis for this limitation in the claim. Claims 1 and 10 recite “training a model at the central node to generate an encoded distribution using a loss function by minimizing a difference between the empirical distribution and the encoded distribution output by the model”. It is unclear how an encoded distribution is generated by itself i.e. the encoded distribution.
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.
Claims 1-18 are rejected under 35 U.S.C. 103 as being unpatentable over Sewell, et. al. (US 2023/0252335 A1; hereinafter, “Sewell”) in view of Hanna, et. al. (O. A. Hanna, Y. H. Ezzeldin, T. Sadjadpour, C. Fragouli and S. Diggavi, "On Distributed Quantization for Classification," in IEEE Journal on Selected Areas in Information Theory, vol. 1, no. 1, pp. 237-249, May 2020, doi: 10.1109/JSAIT.2020.2986467; hereinafter, “Hanna”) in view of Bhorkar, et. al. (US 2020/0371893 A1; hereinafter, “Bhorkar”), and further in view of Shi, et. al. (Y. Shi, K. Yang, T. Jiang, J. Zhang and K. B. Letaief, "Communication-Efficient Edge AI: Algorithms and Systems," in IEEE Communications Surveys & Tutorials, vol. 22, no. 4, pp. 2167-2191, Fourth quarter 2020, doi: 10.1109/COMST.2020.3007787; hereinafter, “Shi”).
Regarding claim 1 and 10, Sewell teaches:
A method comprising: (Sewell, para. 0002: “The present disclosure generally relates to data processing and management. In particular, various embodiments described herein provide systems, methods, techniques, instruction sequences, and devices that use machine learning (ML)-based branching and diving to solve a computational problem that comprises a combinatorial optimization problem”)
A non-transitory storage medium (Sewell, para. 0120: “The terms “machine-readable medium,” “computer-readable medium,” and “device-readable medium” mean the same thing and may be used interchangeably in this disclosure. The terms are defined to include both machine-storage media and transmission media. Thus, the terms include both storage devices/media and carrier waves/modulated data signals. For instance, an embodiment described herein can be implemented using a non-transitory medium (e.g., a non-transitory computer-readable medium).”) having stored therein instructions that are executable by one or more hardware processors to perform operations comprising: (Sewell, para. 0048: “Example methods described herein may also be implemented in the form of executable instructions stored on a machine-readable medium or in the form of electronic circuitry. For instance, the operations of method 300 or method 400 may be represented by executable instructions that, when executed by a digital hardware processor of a computing device, cause the computing device to perform the method 300 or the method 400”)
gathering data from nodes operating in an environment at a central node, (Shi, pg. 2171: “From the system perspective, data distribution (e.g., distributed across edge devices), model parameters (e.g., partitioned and deployed across edge devices and edge servers), computation (e.g., MapReduce), and communication mechanisms (e.g., aggregation at a central node) can be diverse in different applications”) the data including telemetry data; (Shi, pg. 2169: “The connection between data aggregation from distributed nodes in edge training and the in-network computation problem [21] in wireless sensor networks has been established in [22], which proposed an over-the-air computation approach for fast model aggregation in each round of training for on-device federated learning”)
composing, at the central node, inputs from the data (Shi, pg. 2179: “In the distributed system mode, each edge device computes a local update according to its local data samples, and the central node shall periodically aggregate local updates from edge devices”) and a combinatorial optimization problem; (Sewell, para. 0021: “According to some embodiments, ML-based branching and ML-based diving are performed on a combinatorial form (e.g., in the form of Equation 1) of a combinatorial optimization problem”)
solving the combinatorial optimization problem (Sewell, para. 0058: “ The diving algorithm can be used to solve the combinatorial optimization problem (starting with the partial combinatorial solution associated with the identified single node) by dividing the solution space for the combinatorial optimization solution into regions (or dives) that can be explored one by one.”) at the central node (Shi, pg. 2171: “In particular, the full gradient can be computed at a centralized node by aggregating the locally computed partial gradients at all local nodes”)
using the inputs to obtain an optimal solution to generate an empirical distribution of decision variables from the optimal solution; (Hanna, sec. II: “Perhaps the closest approach to ours, are those of learning latent representations for data reconstruction. In variational autoencoders (VAEs) [41], [42], [43], [44], a continuous latent representation space is learned from the inputs, that can then be used to reconstruct inputs or generate new data that follow the same distribution as the data in the training set…” Examiner notes Hanna implicitly teaches generation of a distribution of the optimal solution in order for new generated data to follow such distribution).
training a model at the central node to generate an encoded distribution using a loss function by minimizing a difference between the empirical distribution and the encoded distribution output by the model; and (Hanna, sec. IV(C), Remark 4 and sec. VI(A), Fig. 6: “Remark 4: Note that although dk can take any value in R , only 2N values can make a difference in the misclassification loss in (7): the 2N values corresponding to either coordinate of the training data points {x|(x,y(x))∈T}” “To illustrate the impact of the quantization loss on the distribution of the encoder outputs, Fig. 6 shows the empirical distribution of the encoders outputs after 50 training epochs on the CIFAR-10 dataset,” “Fig. 6. Distribution of the decoder inputs after training for 50 epochs on the CIFAR10 dataset.” Examiner notes that Hanna teaches a central classification model generating an encoded distribution and using a loss function by minimizing a classification loss between empirical distribution and the output of the classification model.)
deploying the model to the nodes from the central node, (Shi, sec. I: “ Specifically, each device only needs to compute a local model according to its own data samples, before sending the computation results to a fusion center, where the global AI model is aggregated and updated. The new AI model will be transmitted back to each device for training at the next epoch”)
wherein the encoded distribution is used to generate a reduced size of the combinatorial optimization problem by fixating a subset of the decision variables. (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints)” Examiner notes Sewell teaches a subset of decision variables being those values that are admissible).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Shi into Sewell. Sewell teaches systems, methods, and non-transitory computer-readable media for using machine learning (ML)-based branching and diving to solve a computational problem that comprises a combinatorial optimization problem; Shi teaches communication-efficient techniques, from both algorithmic and system perspectives for training and inference tasks at a network edge. One of ordinary skill would have been motivated to combine the teachings of Shi into Sewell, as modified, in order to enable more efficient edge AI for coordinating and scheduling edge nodes to efficiently perform a training or inference task under various physical and regulatory constraints (Shi, sec. II(B)).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified. Sewell teaches systems, methods, and non-transitory computer-readable media for using machine learning (ML)-based branching and diving to solve a computational problem that comprises a combinatorial optimization problem. Hanna teaches a pretrained classifier at a central node to carry out its classification on features that are gathered from distributed nodes through communication constrained channels. One of ordinary skill would have been motivated to combine the teachings of Hanna into Sewell as modified in order to design an optimal quantization system which can offer significant processing savings (Hanna, sec. VIII).
Regarding claims 2 and 11, Sewell, as modified, teaches claims 1 and 10 above. Hanna further teaches:
further comprising training the model based on a codification of the combinatorial optimization problem that is input to the model. (Hanna, sec. IV(B): “Given a training dataset T={(x(i),y(i))}Ni=1 , our goal is to design an optimal distributed quantization system (E,D) which minimizes the misclassification loss in (7) for a given communication budget of Rk bits per data point at each node k .”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified, as set forth above with respect to claim 1.
Regarding claims 3 and 12, Sewell, as modified, teaches claims 1 and 10 above. Hanna further teaches:
wherein the inputs are based on information from one or more of the nodes (Hanna, fig. 2: “Fig. 2. An example for distributed quantization of features for classification with K=4 nodes.“)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified, as set forth above with respect to claim 1.
Regarding claims 4 and 13, Sewell, as modified, teaches claims 3 and 12 above. Bhorkar further teaches:
wherein the nodes share an operational context. (Bhorkar, para 0021: “One or more aspects of the subject disclosure include a device comprising a processing system and a memory storing executable instructions that, when executed by the processing system, facilitate performance of operations. The operations comprise receiving data at an edge node of a plurality of edge nodes of a network; the network includes a plurality of regional nodes and a plurality of central nodes. The operations also comprise determining a latency criterion associated with an application for processing the data; the application utilizes an application programming interface (API) included in a database accessible to a central node of the plurality of central nodes (and can also be accessible to edge and regional nodes and to users, as noted above). The operations also comprise measuring and/or accessing a key performance indicator (KPI) of the network, monitoring a latency associated with processing the data by the application, and determining whether the latency satisfies the latency criterion.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Bhorkar into Sewell, as modified. Bhorkar teaches a processing system which receives data at an edge node of a network that also includes regional nodes and central nodes. One of ordinary skill would have been motivated to combine the teachings of Bhorkar into Sewell as modified in order to optimize latency (Bhorkar para. 0038).
Regarding claims 5 and 14, Sewell, as modified, teaches claims 1 and 10 above. Hanna further teaches:
further comprising generating results that include the optimal solution to the combinatorial optimization problem (Hanna, sec. I: “optimal distributed quantization system tailored for classifying a given set of data points.”) at the central node (Hanna, sec. I: “To further reduce complexity and capture richer quantization boundaries (beyond rectangular), we propose a (deep) learning based approach to design our quantizers that makes use of the subdifferentiable nature of the classifier employed by the central node”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified, as set forth above with respect to claim 1.
Regarding claims 6 and 15, Sewell, as modified, teaches claims 1 and 10 above. Hanna further teaches:
further comprising receiving non-feasible solutions from one or more of the nodes. (Hanna, sec. 1: “The communication between the sensors and the central entity comes at a cost (is rate limited), and thus it is expensive to send the measured features with full precision. Instead, each node employs a distributed single-shot quantizer, independently from other nodes, in order to encode its measurements into bit representations that can be sent to the central entity efficiently as soon as sensed. We emphasize that we do not make any a priori distributional assumptions on the data, as is common in many learning scenarios.” Examiner notes that Hanna teaches receiving unfiltered data with no assumptions made from nodes, including both feasible and non-feasible solutions).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified, as set forth above with respect to claim 1.
Regarding claims 7 and 16, Sewell, as modified, teaches claims 6 and 15 above. Hanna and Sewell further teach:
further comprising receiving an input from the one or more of the nodes from which the non-feasible solutions were received, (Hanna, sec. 1: “The communication between the sensors and the central entity comes at a cost (is rate limited), and thus it is expensive to send the measured features with full precision. Instead, each node employs a distributed single-shot quantizer, independently from other nodes, in order to encode its measurements into bit representations that can be sent to the central entity efficiently as soon as sensed. We emphasize that we do not make any a priori distributional assumptions on the data, as is common in many learning scenarios.” Examiner notes that Hanna teaches receiving unfiltered data with no assumptions made from nodes, including both feasible and non-feasible solutions).
wherein the central node (Hanna, sec. I: “To further reduce complexity and capture richer quantization boundaries (beyond rectangular), we propose a (deep) learning based approach to design our quantizers that makes use of the subdifferentiable nature of the classifier employed by the central node”) is configured to determine another optimal solution and return the another optimal solution to the one or more of nodes from which the non-feasible solutions were received. (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints). If the subset at a given node contains only one value, the given node can represent a subproblem that can be solved to determine (e.g., generate) a partial solution to the problem with xi taking the one value. During an iteration of the branch-and-bound algorithm, a leaf node of the branch-and-bound search tree (i.e., a node in the branch-and-bound search tree which currently has no children) can be chosen (e.g., selected) to branch from based on a branching policy implemented by a trained machine learning model, the subproblem represented by the chosen node can be solved as best as possible to determine a partial solution, and a dual bound on an optimal solution of the combinatorial optimization problem can be determined based on the determined partial solution for the chosen node.” Examiner notes “optimal” is defined as “best or most favorable” and therefore for examination purposes only, “another optimal solution” is interpreted as “another valid solution” as taught by Sewell.)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell as set forth above with respect to claim 1.
Regarding claims 8 and 17, Sewell, as modified, teaches claims 1 and 10 above. Sewell further teaches:
The non-transitory storage medium of claim 10, wherein the model enables the nodes to generate the reduced size combinatorial optimization problem that is smaller than a size of the combinatorial optimization problem. (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints). If the subset at a given node contains only one value, the given node can represent a subproblem that can be solved to determine (e.g., generate) a partial solution to the problem with xi taking the one value”)
Regarding claims 9 and 18, Sewell, as modified, teaches claims 8 and 17 above. Sewell further teaches:
The non-transitory storage medium of claim 10, wherein inputs at the nodes are reduced in size by fixating the subset of decision variables, (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints)” Examiner notes Sewell teaches a subset of decision variables being those values that are admissible).
wherein the reduced size combinatorial optimization problem has a smaller search space than a search space of the combinatorial optimization problem. (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints). If the subset at a given node contains only one value, the given node can represent a subproblem that can be solved to determine (e.g., generate) a partial solution to the problem with xi taking the one value”)
Claims 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Sewell, in view of Hanna, in view of Bhorkar, in view of Shi, and further in view of X. Cao, X. Wang and X. Lin ("Design and implementation of a centralized routing protocol for wireless sensor network," 2016 10th International Conference on Sensing Technology (ICST), Nanjing, China, 2016, pp. 1-6, doi: 10.1109/ICSensT.2016.7796227; hereinafter, “Cao”)
Regarding claim 19, Hanna teaches:
A method comprising: receiving a model, at a node, that has been trained at a central node (Shi, sec. I: “ Specifically, each device only needs to compute a local model according to its own data samples, before sending the computation results to a fusion center, where the global AI model is aggregated and updated. The new AI model will be transmitted back to each device for training at the next epoch”), which uses empirical distributions of decision variables derived from an optimal solution of a combinatorial optimization problem and is configured to generate a first encoded distribution of the decision variables, (Hanna, sec. III, algorithm 6: “Ideally, we would like to use an (E,D) system that minimizes the probability of misclassification. That is, the encoders and decoder are the solution of the optimization problem [equation omitted] where: (i) p(x, y(x)) is the input data distribution; (ii)_ y (_x) and _ x are obtained from x using (3) and (5); and in (iii) we used z = E(x) for brevity. However, in this paper we assume that the distribution p(x, y(x)) is not known: instead, we are given a dataset T = {(x(i), y(x(i)))}Ni =1 which contains N independent samples drawn from p(x, y(x)). Thus, we can only empirically approximate the expectation in (6) using the dataset T , and hence, our objective is to minimize the misclassification loss L(E,D, T )” Examiner notes Hanna teaches an empirical approximation of algorithm 6 which includes an input data distribution such that the distribution of the data is an empirical distribution from which an optimal solution is found based on the subsequent calculations provided in algorithms 7 and 8).
wherein the model was trained using the empirical distributions generated from results of the combinatorial optimization problem (Hanna, sec. I: “optimal distributed quantization system tailored for classifying a given set of data points.”) from historical data; (Hanna, sec. IV(C), Remark 4 and sec. VI(A), Fig. 6: “Remark 4: Note that although dk can take any value in R , only 2N values can make a difference in the misclassification loss in (7): the 2N values corresponding to either coordinate of the training data points {x|(x,y(x))∈T}” “To illustrate the impact of the quantization loss on the distribution of the encoder outputs, Fig. 6 shows the empirical distribution of the encoders outputs after 50 training epochs on the CIFAR-10 dataset,” “Fig. 6. Distribution of the decoder inputs after training for 50 epochs on the CIFAR10 dataset.” Examiner notes that Hanna teaches training a model using distributions)
composing an input from data collected at the node; (Hanna, sec. IV(C): “ This can be performed during encoding at each distributed node and reverted in the decoder D at the central node.”)
obtaining a second encoded distribution for the input using the model; (Hanna, sec. II: “Note that the computed zk depends only on x_k , the features available at node k. At the central node, in order to apply the pretrained classifier C, a decoder D generates _ x ∈ Xn from z = [z1, z2, . . . , zK] and uses it as the input to C… We refer to a set of encoders E = {Ek}Kk=1 and a decoder D as a distributed quantization system (E,D).” Examiner notes Hanna teaches a distribution system and then a second quantized distributed system).
generating a reduced input by sampling the second encoded distribution (Hanna, sec. II: “However, in this paper we assume that the distribution p(x, y(x)) is not known: instead, we are given a dataset T = {(x(i), y(x(i)))}Ni =1 which contains N independent samples drawn from p(x, y(x)).”) to fixate a subset of the decision variables, thereby generating a reduced size combinatorial optimization problem; (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints)” Examiner notes Sewell teaches a subset of decision variables being those values that are admissible).
providing the reduced input to the reduced size combinatorial optimization problem to generate a result; (Hanna, sec. V: “The logic behind GBI is as follows. GBI iteratively adds quantization boundaries selected greedily: at each iteration it selects to add one of the possible N boundaries to one of the n features, the one that minimizes the misclassification loss in (8) given the choice of boundaries in the previous iterations.”)
determining whether the result is feasible; (Sewell, para. 0052: “Accordingly, one or more non-root (child) nodes in the branch-and-bound search tree can represent a partial solution to the combinatorial optimization problem with one independent variable, xi, restricted to a subset of admissible values (i.e., values that do not violate any constraints)” Examiner notes for examination reasons only “feasible” is interpreted as possible values i.e. values that do not violate any constraints).
using the result for operations of the node when the result is feasible, (Hanna, sec. V: “GBI iteratively adds quantization boundaries selected greedily: at each iteration it selects to add one of the possible N boundaries to one of the n features, the one that minimizes the misclassification loss in (8) given the choice of boundaries in the previous iterations”)
when the result is non-feasible, transmitting the input to the central node to obtain another optimal solution. (Cao, Sec. II (B): “The central node is a key component in the centralized routing protocol. It gathers network information from all other nodes. Thus, the network diagram with information of all nodes can be set up in the central node. Then, the central node calculates the shortest routing path from each node to its all possible destinations in the graph and disseminates the routing information. As shown in Fig. 2, node A,B and C are sensor nodes and sink node S represents the central node. Operations in the centralized routing protocol can be divided into three phases, information collection, route computation and route maintenance. The main aim of information collection is to establish a network graph with nodes and links between them. Route computation finds the best path between any two nodes in the network. Route maintenance detects the change in the network link state and responds to changes. In Fig. 2, S gathers the information and calculates routing paths while A,B and C just report the local information and receive the route table.” Examiner notes for examination purposes only “non-feasible” is interpreted as not practical where Cao teaches a central node which gathers up information from all other nodes i.e. all other nodes transmit information to the central where the central node then disseminates routing information for a “best path” i.e. an optimal solution).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified, as set forth above with respect to claim 1.
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Cao into Sewell as modified. Cao teaches communication-efficient techniques, from both algorithmic and system perspectives for training and inference tasks at a network edge. One of ordinary skill would have been motivated to combine the teachings of Cao into Sewell, as modified, in order to enable more efficient edge AI for coordinating and scheduling edge nodes to improve the data transmission efficiency and reduce the network loads (Cao sec. I).
Regarding claim 20, Hanna, as modified, teaches claims 19 above. Hanna further teaches:
The method of claim 19, further comprising when the result is non-feasible, receiving the another optimal solution from the central node, (Cao, Sec. II (B): “The central node is a key component in the centralized routing protocol. It gathers network information from all other nodes. Thus, the network diagram with information of all nodes can be set up in the central node. Then, the central node calculates the shortest routing path from each node to its all possible destinations in the graph and disseminates the routing information. As shown in Fig. 2, node A,B and C are sensor nodes and sink node S represents the central node. Operations in the centralized routing protocol can be divided into three phases, information collection, route computation and route maintenance. The main aim of information collection is to establish a network graph with nodes and links between them. Route computation finds the best path between any two nodes in the network. Route maintenance detects the change in the network link state and responds to changes. In Fig. 2, S gathers the information and calculates routing paths while A,B and C just report the local information and receive the route table.” Examiner notes for examination purposes only “non-feasible” is interpreted as not practical where Cao teaches a central node which gathers up information from all other nodes i.e. all other nodes transmit information to the central where the central node then disseminates routing information for a “best path” i.e. an optimal solution).
wherein the central node (Hanna, sec. I: “To further reduce complexity and capture richer quantization boundaries (beyond rectangular), we propose a (deep) learning based approach to design our quantizers that makes use of the subdifferentiable nature of the classifier employed by the central node”) generates the optimal solution by solving the combinatorial optimization problem using the input that generated the non-feasible result at the node. (Hanna, sec. 1: “The communication between the sensors and the central entity comes at a cost (is rate limited), and thus it is expensive to send the measured features with full precision. Instead, each node employs a distributed single-shot quantizer, independently from other nodes, in order to encode its measurements into bit representations that can be sent to the central entity efficiently as soon as sensed. We emphasize that we do not make any a priori distributional assumptions on the data, as is common in many learning scenarios.” Examiner notes that Hanna teaches receiving and using unfiltered data with no assumptions made from nodes, including both feasible and non-feasible input).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Cao into Sewell as modified. Cao teaches communication-efficient techniques, from both algorithmic and system perspectives for training and inference tasks at a network edge. One of ordinary skill would have been motivated to combine the teachings of Cao into Sewell, as modified, in order to enable more efficient edge AI for coordinating and scheduling edge nodes to improve the data transmission efficiency and reduce the network loads (Cao sec. I).
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the teachings of Hanna into Sewell, as modified, as set forth above with respect to claim 1.
Response to Applicant Remarks/Argument
Beginning at the bottom of page 9 of applicant’s remarks, applicant agues that neither Hanna, Bhorkar nor Shi teaches independent claim 1, as amended. However, upon review of the claims as written and the prior art, independent claim 1 nevertheless remains rejected pursuant to 35 USC § 103 for the reasons set forth in the rejection above.
Towards the bottom of page 10 of applicant’s remarks, applicant states that independent claims 10 and 19 recite features similar to claim 1. Both independent claims 10 and 19 likewise remain rejected pursuant to 35 USC § 103 for the reasons set forth in the rejection above.
Applicant further states that dependent claims 2-9, 11-18, and 20 are patentable as a result of their dependence over amended independent claims 1, 10, and 19. Since independent claims remain rejected, unamended dependent claims also remain rejected for the reasons set forth in the rejection above. In particular, independent claim 19 and dependent claim 20 now stands rejected over Sewell in view of Hanna in view of Bhorkar in view of Shi and further in view of Cao.
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 Sally T. Ley whose telephone number is (571)272-3406. The examiner can normally be reached Monday - Thursday, 10:00am - 6:00pm ET.
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, Viker Lamardo can be reached at (571) 270-5871. 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.
/STL/Examiner, Art Unit 2147
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights