CTNF 18/288,415 CTNF 101375 Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. This action is in response to the application and claims filed 10/26/2023. Claims 1, 3-5, 7-9, 11, 13, 15, 18, 20, 26, 28, 31-33, 36, 38 and 45 are pending and have been examined. Claims 1, 3-5, 7-9, 11, 13, 15, 18, 20, 26, 28, 31-33, 36, 38 and 45 are rejected. Information Disclosure Statement The information disclosure statement (IDS) submitted on 10/26/2023 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner. Claim Objections 07-29-01 AIA Claim 8, 32 are objected to because of the following informalities: “wherein the training policy further comprises information that indicates, to the second node whether to up- sample or down-sample”. Applicant may have meant to say “to the first node” not “second node”. The second node established as the actor node that provides the training policy, and the first node is in charge of doing the training of the ML model, therefore, it does not make sense for the training policy to indicate to the second node to whether up-sample or down-sample . Appropriate correction is required. 07-29-01 AIA Claim 11, 18, 36 are objected to because of the following informalities: Claim 11, 18, 36 recites the limitation “the trained ML model”. However, the independent claims that these claims depend on recite “for training a machine learning, ML, model,” . Appropriate correction is required. Claim Rejections - 35 USC § 112 07-30-02 AIA 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. 07-34-01 Claim 32, 38, 45 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention. 07-34-05 AIA Claim 32 recites the limitation " the training dataset " in claim 32 . There is insufficient antecedent basis for this limitation in the claim. 07-34-05 AIA Claim 38 recites the limitation " the other network node " in claim 38 . There is insufficient antecedent basis for this limitation in the claim. Claim 45 recites the limitation "processing circuitry associated with the one or more communication interfaces, the processing circuitry configured to cause the first node " in claim 45. There is insufficient antecedent basis for this limitation in the claim. – EN: Here, the applicant may have meant to state “cause the second node” not first node. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 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. 07-20-aia AIA 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. 07-23-aia AIA The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: 1. Determining the scope and contents of the prior art. 2. Ascertaining the differences between the prior art and the claims at issue. 3. Resolving the level of ordinary skill in the pertinent art. 4. Considering objective evidence present in the application indicating obviousness or nonobviousness. 07-20-02-aia AIA This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention. Examiner’s Note: Some rejections will include an Examiner’s Note (labeled ‘EN’) to provide additional context or rationale explaining the basis for the rejection. 07-21-aia AIA Claim s 1, 3-5, 7, 8, 13, 15, 18, 20 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over Ribera Prat et al. (US 2021/0073675 A1, hereinafter “Ribera”) in view of 3GPP TR 23.700-91 V0.4.0 (“Study on enablers for network automation for the 5G System (5GS); Phase 2,” hereinafter “TR 23.700-91”) . Regarding claim 1, Ribera teaches: “A method performed by a first node for training a machine learning, ML, model, the method comprising:” (Abstract, “A method for training a machine learning model includes: receiving, by a computer system including a processor and memory, a training data” -- EN: this denotes Ribera’s “computer system” as the entity performing the ML method, which corresponds to the “first node” in the claim.) (...) the training policy comprising information that indicates two or more accuracy or importance metrics for two or more ranges of values for a variable to be predicted by the ML model; and (Para 42, “the computer system weights the loss function w(x) …” Para 46, “the weight function w(x) is at 1.75 at values of x from 0.0 to 0.4 … the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7.” -- EN: The training policy is Ribera's weighting scheme, shown by the weight function w(x) that the training process applies across the predicted variable x. That weighting scheme assigns two different values, 1.75 and 0.5, to two different, non-overlapping ranges of the predicted value x ([0.0, 0.4] and [0.4, 0.7]). Each per-range weight is an importance metric that controls how heavily prediction errors in that range are penalized during training, and hence the accuracy the model is to achieve in that range (see Paras 31 and 37: the precision of the model in a portion of the domain depends on the weighting/sampling applied to that portion). The weighting scheme is therefore "information that indicates two or more accuracy or importance metrics for two or more ranges of values for [the] variable to be predicted.") training the ML model based on a training dataset and the training policy. (Para 52, “the computer system trains a machine learning model 372 … based on the training data set 312 using the weighted loss function … to generate a trained model 374” -- EN: Ribera trains the model on the training data set using the weighted loss function, which has the weights/policy, meeting “training … based on a training dataset and the training policy.”) Ribera does not teach: “receiving a training policy for a ML model from a second node,” However, TR 23.700-91 teaches: “receiving a training policy for a ML model from a second node,” (Page 101, cl. 6.24.1.2, step 7c, “Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.” – EN: The Client NWDAF receiving training-influencing "parameters … to help the local model training" from the Server NWDAF (second node) corresponds to "receiving a training policy for a ML model from a second node.”) Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the per-range, label-density-based loss-weighting training technique of Ribera with the training architecture of TR 23.700-91, in which a Server NWDAF (second node) sends parameters “to help the local model training” to a Client NWDAF. The motivation for doing so would be to influence and guide the first node's local training of the ML model, so that the resulting model is trained to produce predictions that are more relevant and useful for that second node's purposes (i.e., accurate in the output ranges on which that node relies) . See TR 23.700-91, cl. 6.24.1.2 (step 7c): “Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.” Regarding claim 3, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 1. Ribera further teaches: wherein training the ML model based on the training dataset and the training policy comprises training the ML model using sample weights applied to samples in the training dataset based on the training policy. (Para 50, “weighting the loss function L(x,x̂) to generate a weighted loss function L_w(x,x̂) is performed by multiplying the loss for any given data point by its corresponding weight” -- EN: multiplying each individual data point’s loss by its own weight corresponds to “sample weights applied to samples … based on the [policy].”) Regarding claim 4, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 3. Ribera further teaches: each sample in the training dataset comprises one or more input variable values and an actual value of the variable to be predicted by the ML model; (Para 53, “The Ames data set in this example includes 1,400 samples, each with 80 features and a sale price.” -- EN: each sample’s “80 features” are the input variable values and the “sale price” is the actual value to be predicted.) the two or more accuracy or importance metrics for the two or more ranges of values for the variable to be predicted by the ML model indicated by the information comprised in the training policy comprise a first accuracy or importance metric for a first range of values for the variable to be predicted by the ML model; (Para 42, “the computer system constructs a weight function w(x) based on the label density f_X(x) and a weighting parameter Δ.” Para 46, “the weight function w(x) is at 1.75 at values of x from 0.0 to 0.4.” -- EN: The weight function w(x) is the “information comprised in the training policy” that indicates a distinct weight value per range; the value 1.75 that w(x) takes on the first range (x from 0.0 to 0.4) is the first accuracy/importance metric for the first range.) the sample weights applied to the samples in the training dataset comprises a first sample weight applied to a first subset of the samples in the training dataset for which the actual value of the variable to be predicted by the ML model is within the first range of values; and (Para 50, “weighting the loss function L(x,x̂) … is performed by multiplying the loss for any given data point by its corresponding weight” Para 46, “the weight function w(x) is at 1.75 at values of x from 0.0 to 0.4” -- EN: each data point in the dataset is multiplied by its own weight, so every sample whose actual predicted value x lies in 0.0–0.4 (the first subset of the dataset) receives the first weight, 1.75.) the first sample weight is based on the first accuracy or importance metric indicated by the information comprised in the training policy for the first range of values. (Para 46 and Para 50 -- EN: the first applied weight (1.75) is the first importance value itself for that range, so the sample weight is “based on” that metric.) Regarding claim 5, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 4. Ribera further teaches: the two or more accuracy or importance metrics for the two or more ranges of values for the variable to be predicted by the ML model indicated by the information comprised in the training policy further comprise a second accuracy or importance metric for second range of values for the variable to be predicted by the ML model, the first and second ranges of values being non-overlapping ranges of values; and (Para 42, “the computer system constructs a weight function w(x) based on the label density f_X(x) and a weighting parameter Δ” Para 46, “the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7” -- EN: The weight function w(x) (the information comprised in the training policy) further indicates the value 0.5 (the second metric) for the second range (x from 0.4 to 0.7), which does not overlap the first range (0.0–0.4).) the sample weights applied to the samples in the training dataset comprises a second sample weight applied to a second subset of the samples in the training dataset for which the actual value of the variable to be predicted by the ML mode is within the second range of values; and (Para 50, “weighting the loss function L(x,x̂) … is performed by multiplying the loss for any given data point by its corresponding weight” Para 46, “the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7” -- EN: each data point in the dataset is multiplied by its own weight, so every sample whose actual predicted value x lies in 0.4–0.7 (the second subset of the dataset) receives the second weight, 0.5.) the second sample weight is based on the second accuracy or importance metric indicated by the information comprised in the training policy for the second range of values. (Para 46, “the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7” and Para 50 -- EN: The second applied weight (0.5) is the same per-range value that w(x) indicates for the second range, so the second sample weight is “based on” the second metric.) Regarding claim 7, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 3. Ribera further teaches: wherein the two or more accuracy or importance metrics for the two or more ranges of values for the variable to be predicted by the ML model are the sample weights. (Para 50, “multiplying the loss for any given data point by its corresponding weight,” these weights being the per-range w(x) values of para 46 (1.75, 0.5) -- EN: in Ribera the per-range importance values (w(x)) are the weights multiplied into the loss, so the metrics “are the sample weights.”) Regarding claim 8, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 1. Ribera further teaches: The method of claim 1 wherein the training policy further comprises information that indicates, to (...), whether to up-sample or down-sample the training dataset for at least one of the two or more ranges of values of the variable to be predicted by the ML model. (Para 39, “One example of a technique used in training classification models from imbalanced training data ( e.g. , where there is a large disparity in the number of samples in different ones of the classes ) includes oversampling ( or duplicating ) data points in the underrepresented classes ( classes with fewer samples ) in the training data and performing the training with this modified data set.” Para 56, “A uniform data set was then generated by randomly selecting one sample from each bin, as shown in the 'after balancing' portion of FIG. 5. This process may be repeated many times to generate a larger uniform testing data set." Para 54, “...the counts (cardinality) of log-scale sale price bins of samples both before and after a balancing procedure... binned into a histogram... this resulted in ten bins.” – EN: Ribera discloses resampling of training data to address imbalance oversampling/duplicating data points in underrepresented portions ([0039]) and, in its FIG. 5 balancing procedure, divides the continuous predicted variable into bins/ranges and resamples per bin, selecting one sample from each bin (thereby down-sampling densely-populated ranges) and repeating the process to grow sparse ranges ([0054]–[0056]). Under BRI, this discloses information indicating whether to up-sample or down-sample the training dataset for at least one of the two or more ranges of values of the predicted variable.) Ribera does not explicitly disclose: “the second node” However, TR 23.700-91 teaches: “the second node” (Page 101, cl. 6.24.1.2, step 7c: "Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.") Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine oversampling (up-/down-sampling) technique of Ribera and apply it to the two metrics as identified in Paragraph 42 of Ribera with the multi-node (Server/Client) federated training architecture of TR 23.700-91, in which a Server NWDAF sends parameters 'to help the local model training' to a client NWDAF. The motivation for doing so would be to enable the first node's training of the ML model to be coordinated and influenced by a separate second node. Thereby allowing the model to be trained without centralizing all of the raw training data in a single node and addressing the data-privacy/security and model-training-efficiency concerns. See TR 23.700-91, cl. 6.24.1.1: 'Federated Learning ... could be a possible solution to handle the issues such as data privacy and security, model training efficiency, and so on, in which there is no need for raw data transferring (e.g. centralized into single NWDAF) but only need for model sharing.” Regarding claim 13, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 1. Ribera further teaches: wherein the first node is a combined training and inferring node, further comprising: (Para 52, “a trained model 374 that includes one or more values that configure the underlying model 372 to compute predictions or inferences…” -- EN: this denotes the first node runs the trained model to compute inferences, thus teaching the combined training and inferring.) generating one or more predicted values for the variable using the ML model; and (...) (Para 35, “a trained predictive model 250, which is trained to compute an estimate of quality 260.”) Ribera does not explicitly teach: “sending the one or more predicted values to the second node.” However, TR 23.700-91 teaches: “sending the one or more predicted values to the second node.” (Page 19, “NWDAF implementations, as per Rel-16 standards, can provide, for a given analytics ID, two different types of output to the consumer of these analytics: either statistics in the past or predictions for the future.” -- EN: 23.700-91’s producing node (NWDAF) transmits its predictions/analytics to the separate consumer (the second node), so that in combination Ribera’s first node, having generated the predictions, sends them to that second node, thus teaching the cross-node transmission that Ribera lacks.) Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the combined training-and-inferring node of Ribera, which generates predicted values using its trained model, with the analytics-delivery operation of TR 23.700-91, by which an analytics-producing node provides its prediction output to the consumer of those analytics. The motivation for doing so would be to deliver the predictions generated by the first node to the separate consumer node on demand, so that the consumer that configured the model’s accuracy can receive and act upon the resulting output. The secondary reference expressly discloses this benefit: See page 19 of TR 23.700-91, cl. 5.2.5.1, “NWDAF implementations, as per Rel-16 standards, can provide, for a given analytics ID, two different types of output to the consumer of these analytics: either statistics in the past or predictions for the future.” Regarding claim 15, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 13. TR 23.700-91 further teaches: sending a model identity, ID, associated to the ML model that is trained based on the training policy to the second node in association with the one or more predicted values. (Page 47, cl. 6.5.1, “The AI Model ID identifies an AI Model and correlates with Analytics ID defined in TS 23.288 [5]... The AIF service consumer provides the following input parameters listed below. A list of AI Model ID(s): identifies requested AI Model(s); … The AIF provides to the consumer the output information listed below. A list of AI Model ID(s); AI Model.” – EN: TR 23.700-91's AI Model ID -- exposed by the model-training AIF (the first node or other network node, which "trains the AI Model(s) and provides AI Model(s) to 5G NFs") to the consumer (second node) and which "identifies an AI Model and correlates with Analytics ID" --is the recited "model identity, ID, associated to the ML model that is trained based on the training policy," received "in association with the one or more predicted values" because those predictions are the analytics delivered to that consumer under the very Analytics ID with which the model ID correlates ) Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the trained model that computes and reports its predicted values of Ribera with the AI Model identification of TR 23.700-91, returned together with the model’s output to identify the specific trained model. The motivation for doing so would be to enable the consumer node to identify which specific model produced the predictions it acts upon and to correlate that model with its analytics output. See TR 23.700-91, clause 6.5.1: “The AI Model ID identifies an AI Model and correlates with Analytics ID defined in TS 23.288.” Regarding claim 18, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 1. Ribera further teaches: wherein the first node is a training node, (Para 52, “the computer system trains a machine learning model 372 … based on the training data set 312 using the weighted loss function … to generate a trained model 374” -- EN: Ribera trains the model on the training data set using the weighted loss function, which has the weights/policy, meeting “training … based on a training dataset and the training policy.”) further comprising sending the trained ML model to an inferring node. (Para 63, “The trained continuous model 374 (e.g., the learned parameters of the model) may then be exported for use, e.g., for deployment and use for performing inferences or predictions in an end user computing device.”) Regarding claim 20, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 18. TR 23.700-91 further teaches: sending a model identity, ID, associated to the ML model that is trained based on the training policy to the second node. (Page 47, cl. 6.5.1, “The AI Model ID identifies an AI Model and correlates with Analytics ID defined in TS 23.288 [5]... The AIF service consumer provides the following input parameters listed below. A list of AI Model ID(s): identifies requested AI Model(s); … The AIF provides to the consumer the output information listed below. A list of AI Model ID(s); AI Model.” – EN: TR 23.700-91's AI Model ID --exposed by the model-training AIF (the first node or other network node, which "trains the AI Model(s) and provides AI Model(s) to 5G NFs") to the consumer and which "identifies an AI Model and correlates with Analytics ID" is the recited "model identity, ID, associated to the ML model that is trained based on the training policy,") Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the trained model that computes and reports its predicted values of Ribera with the AI Model identification of TR 23.700-91, returned together with the model’s output to identify the specific trained model. The motivation for doing so would be to enable the consumer node to identify which specific model produced the predictions it acts upon and to correlate that model with its analytics output. See TR 23.700-91, clause 6.5.1: “The AI Model ID identifies an AI Model and correlates with Analytics ID defined in TS 23.288.” Regarding claim 26, Ribera teaches: “A first node for training a machine learning, ML, model, the first node comprising:” (Abstract, “A method for training a machine learning model includes: receiving, by a computer system including a processor and memory, a training data” -- EN: this denotes Ribera’s “computer system” as the entity performing the ML method, which corresponds to the “first node” in the claim.) one or more communication interfaces; and (Para 62, “The computer system 900 may also include one or more input/output peripherals 908, such as network adapters (e.g., Ethernet and/or WiFi adapters), universal serial bus (USB) adapters, display adapters, and the like.”) processing circuitry associated with the one or more communication interfaces, the processing circuitry configured to cause the first node to: (Para 62, “a computer system 900 includes a processor 902 and memory 904 … The memory 904 stores instructions that, when executed by the processor, cause the processor to implement a method in accordance with embodiments of the present disclosure.”) (...) the training policy comprising information that indicates two or more accuracy or importance metrics for two or more ranges of values for a variable to be predicted by the ML model; and (Para 42, “the computer system weights the loss function w(x) …” Para 46, “the weight function w(x) is at 1.75 at values of x from 0.0 to 0.4 … the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7.” -- EN: The training policy is Ribera's weighting scheme, embodied by the weight function w(x) that the training process applies across the domain of the predicted variable x. That weighting scheme assigns two different values, 1.75 and 0.5, to two different, non-overlapping ranges of the predicted value x ([0.0, 0.4] and [0.4, 0.7]). Each per-range weight is an importance metric that controls how heavily prediction errors in that range are penalized during training, and hence the accuracy the model is driven to achieve in that range (see Paras 31 and 37: the precision of the model in a portion of the domain depends on the weighting/sampling applied to that portion). The weighting scheme is therefore "information that indicates two or more accuracy or importance metrics for two or more ranges of values for [the] variable to be predicted.") training the ML model based on a training dataset and the training policy. (Para 52, “the computer system trains a machine learning model 372 … based on the training data set 312 using the weighted loss function … to generate a trained model 374” -- EN: Ribera trains the model on the training data set using the weighted loss function, which has the weights/policy, meeting “training … based on a training dataset and the training policy.”) Ribera does not teach: “receiving a training policy for a ML model from a second node,” However, TR 23.700-91 teaches: “receiving a training policy for a ML model from a second node,” (Page 101, cl. 6.24.1.2, step 7c, “Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.” – EN: The Client NWDAF receiving training-influencing "parameters … to help the local model training" from the Server NWDAF (second node) corresponds to "receiving a training policy for a ML model from a second node.”) Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the per-range, label-density-based loss-weighting training technique of Ribera with the training architecture of TR 23.700-91, in which a Server NWDAF (second node) sends parameters “to help the local model training” to a Client NWDAF. The motivation for doing so would be to influence and guide the first node's local training of the ML model, so that the resulting model is trained to produce predictions that are more relevant and useful for that second node's purposes (i.e., accurate in the output ranges on which that node relies) . See TR 23.700-91, cl. 6.24.1.2 (step 7c): “Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.” 07-21-aia AIA Claim s 28, 31, 32, 38, and 45 are rejected under 35 U.S.C. 103 as being unpatentable over 3GPP TR 23.700-91 V0.4.0 (“Study on enablers for network automation for the 5G System (5GS); Phase 2,” hereinafter “TR 23.700-91”) in view of Ribera Prat et al. (US 2021/0073675 A1, hereinafter “Ribera”) Regarding claim 28, TR 23.700-91 teaches: A method performed by a second node for influencing training of a machine learning, ML, model, the method comprising: (Page 99-100, “...build machine-learning models based on data sets that are distributed in different network functions. A Client NWDAF (e.g. deployed in a domain or network function) locally trains the local ML model with its own data and share it to the server NWDAF. With local ML models from different Client NWDAFs, the Server NWDAF could aggregate them into a global or optimal ML model or ML model parameters and send them back to the Client NWDAFs for inference.” – EN: The Server NWDAF, which directs and influences the training carried out at the client(s), corresponds to the recited “second node … for influencing training.”) sending a training policy for a ML model to a first node, (...) (Page 101, cl. 6.24.1.2, step 7c, “Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.” – EN: The Server NWDAF (second node) sending the training-influencing parameters to the Client NWDAF corresponds to “sending a training policy … to a first node,”) receiving one or more predicted values for the variable to be predicted by the ML model from either the first node or another node. (Page 19, cl. 5.2.5.1, “NWDAF implementations, as per Rel-16 standards, can provide, for a given analytics ID, two different types of output to the consumer of these analytics: either statistics in the past or predictions for the future.” – EN: The second node's receipt of the producing NWDAF's "predictions for the future" (cl. 5.2.5.1) is the recited "receiving one or more predicted values … for the variable to be predicted,") TR 23.700-91 does not explicitly teach: the training policy comprising information that indicates two or more accuracy or importance metrics for two or more ranges of values for a variable to be predicted by the ML model; and However, Ribera teaches: the training policy comprising information that indicates two or more accuracy or importance metrics for two or more ranges of values for a variable to be predicted by the ML model; and (Para 42, “the computer system weights the loss function w(x) …” Para 46, “the weight function w(x) is at 1.75 at values of x from 0.0 to 0.4 … the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7.” -- EN: The training policy is Ribera's weighting scheme, embodied by the weight function w(x) that the training process applies across the domain of the predicted variable x. That weighting scheme assigns two different values, 1.75 and 0.5, to two different, non-overlapping ranges of the predicted value x ([0.0, 0.4] and [0.4, 0.7]). Each per-range weight is an importance metric that controls how heavily prediction errors in that range are penalized during training, and hence the accuracy the model is driven to achieve in that range (see Paras 31 and 37: the precision of the model in a portion of the domain depends on the weighting/sampling applied to that portion). The weighting scheme is therefore "information that indicates two or more accuracy or importance metrics for two or more ranges of values for [the] variable to be predicted.") Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the training architecture of TR 23.700-91, in which a Server NWDAF (second node) sends parameters “to help the local model training” to a Client NWDAF with the per-range, label-density-based loss-weighting training technique of Ribera The motivation for doing so would be to improve the accuracy of the trained model when its training data is imbalanced. In particular, to raise the model's accuracy in the sparse, under-represented ranges of the predicted variable that an unweighted loss would otherwise leave inaccurate. See Ribera, Para 2: "Aspects of embodiments of the present disclosure relate to a system and method to improve the accuracy of regression models trained with imbalanced data." Regarding claim 31, as discussed above, TR 23.700-91 in view of Ribera teaches all of the limitations of claim 28. Ribera further teaches: wherein the two or more accuracy or importance metrics for the two or more ranges of values for the variable to be predicted by the ML model are sample weights to be used for training the ML model. (Para 50, "multiplying the loss for any given data point by its corresponding weight," these weights being the per-range w(x) values of para 46 (1.75, 0.5) -- EN: in Ribera the per-range importance values (1.75, 0.5) are the very weights multiplied into the loss during training, so the metrics "are sample weights to be used for training the ML model.") Regarding claim 32, as discussed above, TR 23.700-91 in view of Ribera teaches all of the limitations of claim 28. TR 23.700-91 teaches: “the second node” (Page 101, cl. 6.24.1.2, step 7c: "Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning.") TR 23.700-91 does not explicitly teach: The method of claim 28 wherein the training policy further comprises information that indicates, (...), whether to up-sample or down-sample the training dataset for at least one of the two or more ranges of values of the variable to be predicted by the ML model. However, Ribera teaches: The method of claim 28 wherein the training policy further comprises information that indicates, (...), whether to up-sample or down-sample the training dataset for at least one of the two or more ranges of values of the variable to be predicted by the ML model. (Para 39, “One example of a technique used in training classification models from imbalanced training data ( e.g. , where there is a large disparity in the number of samples in different ones of the classes ) includes oversampling ( or duplicating ) data points in the underrepresented classes ( classes with fewer samples ) in the training data and performing the training with this modified data set.” Para 56, “A uniform data set was then generated by randomly selecting one sample from each bin, as shown in the 'after balancing' portion of FIG. 5. This process may be repeated many times to generate a larger uniform testing data set." Para 54, “...the counts (cardinality) of log-scale sale price bins of samples both before and after a balancing procedure... binned into a histogram... this resulted in ten bins.” – EN: Ribera discloses resampling of training data to address imbalance oversampling/duplicating data points in underrepresented portions ([0039]) and, in its FIG. 5 balancing procedure, divides the continuous predicted variable into bins/ranges and resamples per bin, selecting one sample from each bin (thereby down-sampling densely-populated ranges) and repeating the process to grow sparse ranges ([0054]–[0056]). Under BRI, this discloses information indicating whether to up-sample or down-sample the training dataset for at least one of the two or more ranges of values of the predicted variable.) Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the multi-node (Server/Client) federated training architecture of TR 23.700-91 in which a Server NWDAF (second node) sends parameters 'to help the local model training' to a Client NWDAF with the up-/down-sampling (resampling/balancing) technique of Ribera and apply it to the two metrics as identified in Paragraph 42 of Ribera. The motivation for doing so would be to improve the accuracy of the resulting ML model's predictions in the under-represented ranges of the predicted variable (the ranges having fewer training samples) by rebalancing the training dataset so those sparse ranges are not dominated by densely-populated ranges during training. See Ribera [0031]: 'The imbalance in the data distribution can impact the accuracy of the predictions made in that portion... the model may increase the accuracy in the denser parts of the data set at the expense of reduced accuracy in the sparser parts of the data set.'" Regarding claim 38, as discussed above, TR 23.700-91 in view of Ribera teaches all of the limitations of claim 28. TR 23.700-91 further teaches: receiving a model identity, ID, associated to the ML model that is trained based on the training policy from the first node or the other network node, in association with the one or more predicted values. (Page 47, cl. 6.5.1, “The AI Model ID identifies an AI Model and correlates with Analytics ID defined in TS 23.288 [5]... The AIF service consumer provides the following input parameters listed below. A list of AI Model ID(s): identifies requested AI Model(s); … The AIF provides to the consumer the output information listed below. A list of AI Model ID(s); AI Model.” – EN: TR 23.700-91's AI Model ID --exposed by the model-training AIF (the first node or other network node, which "trains the AI Model(s) and provides AI Model(s) to 5G NFs") to the consumer (second node) and which "identifies an AI Model and correlates with Analytics ID" --is the recited "model identity, ID, associated to the ML model that is trained based on the training policy," received "in association with the one or more predicted values" because those predictions are the analytics delivered to that consumer under the very Analytics ID with which the model ID correlates ) Regarding claim 45, TR 23.700-91 teaches: A second node for influencing training of a machine learning, ML, model, the second node comprising: (...) (Page 99-100, "…build machine-learning models based on data sets that are distributed in different network functions. A Client NWDAF (e.g. deployed in a domain or network function) locally trains the local ML model with its own data and share it to the server NWDAF. With local ML models from different Client NWDAFs, the Server NWDAF could aggregate them into a global or optimal ML model or ML model parameters and send them back to the Client NWDAFs for inference." – EN: The Server NWDAF, which directs and influences the training carried out at the client(s), corresponds to the recited "second node … for influencing training.") send a training policy for a ML model to a first node, (...) (Page 101, cl. 6.24.1.2, step 7c, "Server NWDAF sends analytics request to the Client NWDAFs that participate in the Federated model according to steps 7a and 7b including some parameters (such as data type list, maximum response time window, etc.) to help the local model training for Federated Learning." – EN: The Server NWDAF (second node) sending the training-influencing parameters to the Client NWDAF corresponds to "send a training policy … to a first node.") receive one or more predicted values for the variable to be predicted by the ML model from either the first node or another node. (Page 19, cl. 5.2.5.1, "NWDAF implementations, as per Rel-16 standards, can provide, for a given analytics ID, two different types of output to the consumer of these analytics: either statistics in the past or predictions for the future." – EN: The second node's receipt of the producing NWDAF's "predictions for the future" (cl. 5.2.5.1) is the recited "receive one or more predicted values … for the variable to be predicted.") TR 23.700-91 does not explicitly teach: one or more communication interfaces comprising either or both of: (i) a network interface and (ii) one or more radio units; and processing circuitry associated with the one or more communication interfaces, the processing circuitry configured to cause the [second] node to: the training policy comprising information that indicates two or more accuracy or importance metrics for two or more ranges of values for a variable to be predicted by the ML model; and However, Ribera teaches: one or more communication interfaces comprising either or both of: (i) a network interface and (ii) one or more radio units; and (Para 62, "The computer system 900 may also include one or more input/output peripherals 908, such as network adapters (e.g., Ethernet and/or WiFi adapters), universal serial bus (USB) adapters, display adapters, and the like.") processing circuitry associated with the one or more communication interfaces, the processing circuitry configured to cause the [second] node to: (Para 62, "a computer system 900 includes a processor 902 and memory 904 … The memory 904 stores instructions that, when executed by the processor, cause the processor to implement a method in accordance with embodiments of the present disclosure.") the training policy comprising information that indicates two or more accuracy or importance metrics for two or more ranges of values for a variable to be predicted by the ML model; and (Para 42, "the computer system weights the loss function w(x) …" Para 46, "the weight function w(x) is at 1.75 at values of x from 0.0 to 0.4 … the weight function w(x) drops to a minimum of 0.5 between values of x from 0.4 to 0.7." -- EN: The training policy is Ribera's weighting scheme, embodied by the weight function w(x) that the training process applies across the domain of the predicted variable x. That weighting scheme assigns two different values, 1.75 and 0.5, to two different, non-overlapping ranges of the predicted value x ([0.0, 0.4] and [0.4, 0.7]). Each per-range weight is an importance metric that controls how heavily prediction errors in that range are penalized during training, and hence the accuracy the model is driven to achieve in that range (see Paras 31 and 37: the precision of the model in a portion of the domain depends on the weighting/sampling applied to that portion). The weighting scheme is therefore "information that indicates two or more accuracy or importance metrics for two or more ranges of values for [the] variable to be predicted.") Claim 45 is the second-node apparatus counterpart of method claim 28. Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the training architecture of TR 23.700-91, in which a Server NWDAF (second node) sends parameters "to help the local model training" to a Client NWDAF, with the per-range, label-density-based loss-weighting training technique and the computer system hardware of Ribera. The motivation for doing so would be to provide a second-node apparatus whose processing circuitry can influence the producing node's training and receive the resulting predictions, thereby improving the accuracy of the trained model when its training data is imbalanced. See Ribera, Para 2: "Aspects of embodiments of the present disclosure relate to a system and method to improve the accuracy of regression models trained with imbalanced data." 07-21-aia AIA Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Ribera Prat et al. (US 2021/0073675 A1, hereinafter “Ribera”) in view of 3GPP TR 23.700-91 V0.4.0 (“Study on enablers for network automation for the 5G System (5GS); Phase 2,” hereinafter “TR 23.700-91”) and further in view of Duan et al. (“Astraea: Self-balancing Federated Learning for Improving Classification Accuracy of Mobile Deep Learning Applications,” hereinafter “Duan”) . Regarding claim 9, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 1. Duan teaches: further comprising, prior to receiving the training policy from the second node, (page 5, and Fig. 3, "…the FL server counts the global data distribution and initializes the weights and the optimizer of the learning model. … In the rebalancing phase, the server first calculates the amount of augmentations for each class based on the global data distribution. Then, all clients perform data augment in parallel according to the calculation results (②)." -- EN: Per the Astraea workflow, which is ordered Initialization → Rebalancing → Training (Fig. 3), the clients (first node) transmit their distribution information during initialization, and the server-determined rebalancing/policy is calculated and applied only in the later rebalancing phase. The dataset information is therefore sent before the training node receives the policy, which corresponds to "prior to receiving the training policy from the second node.") sending information about the training dataset to the second node. (Page 5, "The devices participate in the training by sending their local data distribution information to the server (①)." -- EN: Duan's clients are the first node (the nodes that train the ML model on their local data); the FL server is the second node. The clients' "local data distribution information" characterizes the composition of their local training data, i.e., "information about the training dataset.") Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the policy-configured ML training of Ribera and the training architecture of TR 23.700-91 with the self-balancing federated-learning step of Duan, in which the training clients report their local data-distribution information to the server before the server determines and applies the rebalancing policy. The motivation for doing so would be to allow the policy to be tailored to the actual composition of the training data, thereby alleviating data imbalance and improving classification accuracy. The third reference expressly discloses this benefit: Duan, Abstract: "Astraea shows +5.59% and +5.89% improvement of top-1 accuracy on the imbalanced EMNIST and imbalanced CINIC-10 datasets, respectively." 07-21-aia AIA Claim 33 is rejected under 35 U.S.C. 103 as being unpatentable over 3GPP TR 23.700-91 V0.4.0 (“Study on enablers for network automation for the 5G System (5GS); Phase 2,” hereinafter “TR 23.700-91”) in view of Ribera Prat et al. (US 2021/0073675 A1, hereinafter “Ribera”) further in view of Duan et al. (“Astraea: Self-balancing Federated Learning for Improving Classification Accuracy of Mobile Deep Learning Applications,” hereinafter “Duan”) . Regarding claim 33, as discussed above, TR 23.700-91 in view of Ribera teaches all of the limitations of claim 28. Duan teaches: The method of claim 28 further comprises determining the training policy, further comprising: receiving, from the first node, information about a training dataset to be used at the first node to train the ML model; and (page 5, "The devices participate in the training by sending their local data distribution information to the server (①). After determining the devices (clients) to be involved in the training, the FL server counts the global data distribution…"; see also page 2, "each device trains model using its local data." -- EN: The FL server receives the local data distribution information from the clients. Because each client uses that local data to train the model, the reported distribution is "information about a training dataset to be used at the first node to train the ML model.") wherein determining the training policy comprises determining the training policy based on the information about the training dataset. (Page 5 and Algorithm 2 (FL Server), "the server first calculates the amount of augmentations for each class based on the global data distribution"; "Calculate the data size of each class C₁,…,Cₙ, and the mean C̄. for each class i in 1,…,N do if Cᵢ < C̄ then Augmentation set Y_aug ∪ i." -- EN: The FL server (second node) computes the rebalancing/policy from the global data distribution, which is the aggregate of the per-client distribution information received at step ① . Determining the policy from the reported distribution corresponds to "determining the training policy based on the information about the training dataset.") Claim 33 is the second-node counterpart of claim 9. Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the training architecture of TR 23.700-91 and the per-range loss-weighting technique of Ribera with the FL server of Duan, which receives the clients' local data-distribution information and determines the rebalancing policy from it. The motivation for doing so would be to allow the second node to tailor the training policy to the actual composition of the reported training data, thereby alleviating data imbalance and improving classification accuracy. The third reference expressly discloses this benefit: Duan, Abstract: "Astraea shows +5.59% and +5.89% improvement of top-1 accuracy on the imbalanced EMNIST and imbalanced CINIC-10 datasets, respectively." 07-21-aia AIA Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Ribera Prat et al. (US 2021/0073675 A1, hereinafter “Ribera”) in view of 3GPP TR 23.700-91 V0.4.0 (“Study on enablers for network automation for the 5G System (5GS); Phase 2,” hereinafter “TR 23.700-91”), and further in view of Wang et al. (“Addressing Class Imbalance in Federated Learning,” hereinafter “Wang”) . Regarding claim 11, as discussed above, Ribera in view of TR 23.700-91 teaches all of the limitations of claim 1 TR 23.700-91 further teaches: sending information about the trained ML model to the second node; (page 101, clause 6.24.1.2, step 9: "each Client NWDAF trains local ML model based on its own data and the parameters received from Server NWDAF, and reports the local ML model information (e.g., volume of the local dataset, parameters of the local ML model) to the Server NWDAF." – EN: The Client NWDAF reporting "local ML model information" including "parameters of the local ML model" to the Server NWDAF (second node) corresponds to "sending information about the trained ML model to the second node." Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the per-range importance-weighted loss function training of Ribera and the multi-node architecture and the server-to-client training parameter of TR 23.700-91. The motivation for doing so would be to deliver the predictions generated by the first node to the separate consumer node on demand, so that the consumer that configured the model’s accuracy can receive and act upon the resulting output. The secondary reference expressly discloses this benefit: See page 19 of TR 23.700-91, cl. 5.2.5.1, “NWDAF implementations, as per Rel-16 standards, can provide, for a given analytics ID, two different types of output to the consumer of these analytics: either statistics in the past or predictions for the future.” Ribera in view of TR 23.700-91 does not explicitly teach: receiving an updated training policy from the second node; and updating or re-training the ML model based on the updated training policy. However, Wang teaches: receiving an updated training policy from the second node; and (Page 5, "Once our monitor detects a similar imbalanced composition continuously by checking v_pt, it will acknowledge that the global model has learned imbalanced data and apply a mitigation strategy that is based on our Ratio Loss function. … After computing all Ra_{p,i} for class p as Eq. (6), we can get their average value and its corresponding absolute value, denoted as Ra_p, and compose R = [Ra_1, ..., Ra_p, ..., Ra_Q]. Finally, in the local training, when a sample X^{(p)} of class p is fed to the neural network, its corresponding loss is: L_{RL}(X^{(p)}) = −(α + βRa_p) · log(f^{(p)}_{(p)}) … We mitigate the impact of class imbalance by modifying the coefficient π before L_{CE}. When the input is a minority class, according to Theorem 2, its corresponding Ra is relatively large, and then its contribution to the overall loss will increase, and vice versa. – EN: The server-side monitor evaluates the trained global model each round, recomputes updated per-class importance weights R = [Ra₁, …, Ra_Q], and provides them to the clients for subsequent local training, which corresponds to "receiving an updated training policy from the second node.") updating or re-training the ML model based on the updated training policy. (Page 5, “Finally, in the local training, when a sample X(p) of class p is fed to the neural network, its corresponding loss is: (See equation 11)” – The clients perform local training using the Ratio Loss function configured with the updated per-class importance weights R received from the server/monitor, which corresponds to "re-training the ML model based on the updated training policy.") Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the per-range importance-weighted loss function training of Ribera and the multi-node architecture and the server-to-client training parameter of TR 23.700-91 with the monitoring and updated per-class importance-metric feedback loop of Wang. The motivation for doing so would be to enable the server node to detect and correct data imbalance across successive training rounds, thereby preventing accuracy degradation in underrepresented output regions and improving overall model performance. See Wang, page 2: "If such imbalance cannot be detected in time, it will induce the common model to the wrong direction in the early training phase, and thus poison the common model and deteriorate the performance." 07-21-aia AIA Claim 36 is rejected under 35 U.S.C. 103 as being unpatentable over 3GPP TR 23.700-91 V0.4.0 (“Study on enablers for network automation for the 5G System (5GS); Phase 2,” hereinafter “TR 23.700-91”) in view of Ribera Prat et al. (US 2021/0073675 A1, hereinafter “Ribera”), and further in view of Wang et al. (“Addressing Class Imbalance in Federated Learning,” hereinafter “Wang”) . Regarding claim 36, as discussed above, TR 23.700-91 in view of Ribera teaches all of the limitations of claim 28. TR 23.700-91 further teaches: receiving information about the trained ML model from the first node; (Page 101, cl. 6.24.1.2, "each Client NWDAF trains local ML model based on its own data and the parameters received from Server NWDAF, and reports the local ML model information (e.g., volume of the local dataset, parameters of the local ML model) to the Server NWDAF." --EN: The Server NWDAF receiving "local ML model information" from the Client NWDAF corresponds to "receiving information about the trained ML model from the first node.") Ribera in view of TR 23.700-91 does not explicitly teach: determining an updated training policy based on the information about the trained ML model; and sending the updated training policy to the first node. However, Wang teaches: determining an updated training policy based on the information about the trained ML model; and (Page 5, "Once our monitor detects a similar imbalanced composition continuously by checking v_pt, it will acknowledge that the global model has learned imbalanced data and apply a mitigation strategy that is based on our Ratio Loss function."; Page 5, "After computing all Ra_{p,i} for class p as Eq. (6), we can get their average value and its corresponding absolute value, denoted as Ra_p, and compose R = [Ra_1, ..., Ra_p, ..., Ra_Q]." --EN: The server-side monitor evaluates the global model, which incorporates the trained model information received from clients and computes updated per-class importance weights R, which corresponds to "determining an updated training policy based on the information about the trained ML model.") sending the updated training policy to the first node. (Page 5, "When the input is a minority class, according to Theorem 2, its corresponding Ra is relatively large, and then its contribution to the overall loss will increase, and vice versa."; Page 2, "the system will acknowledge that the global model has learned imbalanced data, and then try to mitigate its impact by applying the Ratio Loss in FL." --EN: The server-side monitor provides the recomputed per-class importance weights R to the clients for use in subsequent local training rounds, which corresponds to "sending the updated training policy to the first node.") Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the federated ML training architecture of TR 23.700-91 and the per-range importance-weighted training technique of Ribera with the iterative monitoring and updated per-class importance-metric feedback loop of Wang. The motivation for doing so would be to enable the server node to detect and correct data imbalance across successive federated training rounds, thereby preventing accuracy degradation in underrepresented output regions and improving overall model performance. See Wang, page 2: "If such imbalance cannot be detected in time, it will induce the common model to the wrong direction in the early training phase, and thus poison the common model and deteriorate the performance." Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to NAYMUR RAHMAN ALI whose telephone number is (571)272-0007. The examiner’s email address is Naymur.ali@uspto.gov . The examiner can normally be reached Mon-Fri. 9:30-6:30 pm. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alexey Shmatov can be reached at (571)270-3428. 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. /NAYMUR RAHMAN ALI/Examiner, Art Unit 2123 /ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123 Application/Control Number: 18/288,415 Page 2 Art Unit: 2123 Application/Control Number: 18/288,415 Page 3 Art Unit: 2123 Application/Control Number: 18/288,415 Page 4 Art Unit: 2123 Application/Control Number: 18/288,415 Page 5 Art Unit: 2123 Application/Control Number: 18/288,415 Page 6 Art Unit: 2123 Application/Control Number: 18/288,415 Page 7 Art Unit: 2123 Application/Control Number: 18/288,415 Page 8 Art Unit: 2123 Application/Control Number: 18/288,415 Page 9 Art Unit: 2123 Application/Control Number: 18/288,415 Page 10 Art Unit: 2123 Application/Control Number: 18/288,415 Page 11 Art Unit: 2123 Application/Control Number: 18/288,415 Page 12 Art Unit: 2123 Application/Control Number: 18/288,415 Page 13 Art Unit: 2123 Application/Control Number: 18/288,415 Page 14 Art Unit: 2123 Application/Control Number: 18/288,415 Page 15 Art Unit: 2123 Application/Control Number: 18/288,415 Page 16 Art Unit: 2123 Application/Control Number: 18/288,415 Page 17 Art Unit: 2123 Application/Control Number: 18/288,415 Page 18 Art Unit: 2123 Application/Control Number: 18/288,415 Page 19 Art Unit: 2123 Application/Control Number: 18/288,415 Page 20 Art Unit: 2123 Application/Control Number: 18/288,415 Page 21 Art Unit: 2123 Application/Control Number: 18/288,415 Page 22 Art Unit: 2123 Application/Control Number: 18/288,415 Page 23 Art Unit: 2123 Application/Control Number: 18/288,415 Page 24 Art Unit: 2123 Application/Control Number: 18/288,415 Page 25 Art Unit: 2123 Application/Control Number: 18/288,415 Page 26 Art Unit: 2123 Application/Control Number: 18/288,415 Page 27 Art Unit: 2123 Application/Control Number: 18/288,415 Page 28 Art Unit: 2123 Application/Control Number: 18/288,415 Page 29 Art Unit: 2123 Application/Control Number: 18/288,415 Page 30 Art Unit: 2123 Application/Control Number: 18/288,415 Page 31 Art Unit: 2123 Application/Control Number: 18/288,415 Page 32 Art Unit: 2123 Application/Control Number: 18/288,415 Page 33 Art Unit: 2123
Prosecution Insights