Prosecution Insights
Last updated: November 20, 2026
Application No. 18/276,972

ENERGY-EFFICIENT DEEP NEURAL NETWORK TRAINING ON DISTRIBUTED SPLIT ATTRIBUTES

Final Rejection §101§103
Filed
Aug 11, 2023
Priority
Feb 15, 2021 — provisional 63/149,474 +1 more
Examiner
NGUYEN, TRI T
Art Unit
2128
Tech Center
2100 — Computer Architecture & Software
Assignee
Telefonaktiebolaget LM Ericsson
OA Round
2 (Final)
67%
Grant Probability
Favorable
3-4
OA Rounds
8m
Est. Remaining
82%
With Interview

Examiner Intelligence

Grants 67% — above average
67%
Career Allowance Rate
136 granted / 203 resolved
+12.0% vs TC avg
Moderate +14% lift
Without
With
+14.5%
Interview Lift
resolved cases with interview
Typical timeline
3y 12m
Avg Prosecution
17 currently pending
Career history
228
Total Applications
across all art units

Statute-Specific Performance

§101
16.3%
-23.7% vs TC avg
§103
61.5%
+21.5% vs TC avg
§102
3.2%
-36.8% vs TC avg
§112
15.2%
-24.8% vs TC avg
Black line = Tech Center average estimate • Based on career data from 203 resolved cases

Office Action

§101 §103
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 . Response to Amendment The amendment filed 08/12/2026 has been entered. Claims 1-19 and 22 remain pending in the application. Response to Arguments Applicant’s arguments, filed 08/12/2026, with respect to the rejections of the claims under 101 have been fully considered and are not persuasive. Applicant argues (page 3) The Office Action states that the limitations of "determining whether layer outputs for the cut-layer were not received from one of the workers," "in response to determining that the layer outputs for the cut-layer were not received from one of the workers," and "generating imputed values of the layer outputs that were not received," and "splitting the gradients into groups associated with respective ones of the workers" under its broadest reasonable interpretation covers performance of the limitation in the mind. (See Office Action, p. 3). The Office Action further states that "calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs" recites a mathematical concept. However, independent Claim 1 (and similarly in Claim 22) specify that the method is for operating a master node which is a network device. Moreover, the master node is receiving information from workers which are defined as network nodes such as a base station, UE, serving gateway, or a packet gateway. Thus, Claim 1 is tied to particular hardware under which the broadest reasonable interpretation of the claims would not cover performance of the limitations in the mind. In response The examiner respectfully disagrees. The claim does not define a master node as a network device. Also, the claim does not recite how the master node is used/operated to perform the claim steps in such a specific way such that the human mind cannot be mentally performed. As stated in the 101 rejections section of the Office Action mailed on 05/13/2026, the claim recites the steps that can be performed in a human mind. For example, a user can determine if data is received from an entity, if not, assign a value for the missing data based on the past received data for example, and the user can group data together based on certain factor. If a claim, under the broadest reasonable interpretation covers concepts that can be performed in the human mind, or by a human using a pen and paper, including observation, evaluation, judgment, or opinion, it will be considered as falling within the "mental process” grouping of abstract idea. Applicant argues (pages 3-5) Applicant respectfully submits that these features of Claim 1 integrate the alleged judicial exception into a practical application for the reasons discussed below. Step 2A, Prong Two Applicant does not concede that Claim 1 recites the purported abstract idea of mental process. However, Applicant respectfully submits that Claim 1 is eligible under 35 U.S.C. 101 even if Claim 1 recites a mental process, at least, because when Claim 1 is evaluated as a whole, including viewing the additional claim elements individually and in combination, the purported mental process is integrated into a practical application. The Office Action states that the additional limitations of "a master node," "a vertical federated learning, vFL system," "a split neural network" and "a plurality of workers" are recited at a high-level of generality (i.e., as a generic device performing the generic computer functions) such that they amount to no more than mere instructions to apply the exception using the generic computer components (MPEP 2106.05(f)). Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See Office Action, p. 4). However, Claim 1 specifies detailed steps of non-generic computer functions performed by the master node based on data received from the plurality of workers in a split neural network. Contrary, to the assertion made in the Office Action, Claim 1 amounts to more than "mere instructions to apply the exception using the generic computer components." For example, Claim 1 provides an improvement to vertical federated learning (vFL) solutions that function via split neural networks. For example, functional requirements for vFL are more restrictive than horizontal federated learning. For example, vFL requires that all parties need to participate in every round at every round. If one party does not participate the accuracy of the predictions made by the split neural network may be negatively affected. (See e.g., Specification, para. [0074]). The present Claim 1 (and similarly Claim 22) provides a method that addresses the loss of one or more workers due to a failure in a communication link or other failure that prevents them from transmitting/receiving data (e.g., participating in a round). For example, some embodiments provide a method that imputes missing data with minimal loss to accuracy. (See e.g., Specification, para. [0077]-[0083]). Various improvements are reflected in the claim language when taken as a whole, including viewing the additional claim elements individually and in combination, as the claim recites amongst other things, "in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received." For example, by generating imputed values of the layer outputs that were not received, the claim recitations reflect improvements including increasing the adaptability and/or flexibility of the split neural network without sacrificing accuracy. Accordingly, Applicant respectfully submits that under Step 2A Prong Two, Claim 1 (and similarly Claim 22) is patent eligible under 35 U.S.C. 101 because the claim language recite an improvement in computer functionality and improvement in the field of technology such that the purported mental process is integrated into a practical application. As such, withdrawal of the rejection of Claim 1 (and similarly Claim 22) under 35 U.S.C. 101 is respectfully requested. In response The examiner respectfully disagrees. Claim 1 in the preamble recites a method … for training a split neural network, however, the claim does not clearly show how the split neural network is trained and/or updated. The claim reciting steps of (among others) “receiving layer outputs … generating imputed values of the layer outputs that were not received … calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs”. These steps only show how the gradient is calculated using the outputs from the clients and the imputed data for the missing outputs, but as stated in the 101 rejections section mailed on 05/13/2026, the process of “calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs” recites a mathematical concept (group of abstract idea), thus, at best, the claim recites the steps that maybe improve the “calculating gradients” using the outputs from the clients and the imputed data for the missing outputs, however, an improvement in an abstract idea is not consider an improvement on the functioning of a computer or to any other technology or technical field. The claim does not clearly show how the split neural network is trained using the imputed data, and why generating the imputed data for the missing outputs is an improvement on the functioning of a computer or to any other technology or technical field. Indeed, the claim does not recite any training step that train the split neural network. Further, the additional elements of “receiving layer outputs for a sample period from one or more of the workers for a cut-layer at which the neural network is split between the workers and the master node” and “transmitting the groups of gradients to respective ones of the workers” amount to insignificant extra-solution activities of data gathering and transmitting which do not amount to significantly more than the abstract idea (MPEP 2106.05(g)). Therefore, even when considering the additional elements in combination, the claim as a whole does not integrate the recited exception into a practical application. Applicant argues (pages 5-6) Step 2B Applicant does not concede that Claim 1 recites a mental process that is not integrated into a practical application. However, Applicant respectfully submits that the recitations of the claims amount to significantly more than the abstract idea of mental processes as stated by the Office Action. The Federal Circuit noted further in Bascom: … if the subject matter is patentable, it is not an abstract idea. (Bascom concurrence at 6, J. Newman, concurring, emphasis added.) Because, as shown below in the Remarks, the pending claims are patentable over the art of record and thus meet the requirements of novelty and non-obviousness of 35 U.S.C. §§ 102 and 103, the claims include unconventional limitations and unconventional steps other than what is well understood, routine, and conventional, and these limitations qualify as "significantly more" than any judicial exception. Accordingly, Applicant respectfully asserts that the recitations of the pending claims are not directed to an abstract idea and, thus, do not meet the first test for the judicial exception. Applicant also asserts that even if, accepting only for the sake of the current argument, the claims were directed to an abstract idea and deemed not to be integrated into a practical application, their recitations add significantly more to the abstract idea. In response The examiner respectfully disagrees. The examiner has explained above and in the 101 rejections section why the claim recites an abstract idea and why the claim as a whole does not integrate the recited exception into a practical application. Also, in the 101 rejections section, the rejections identified additional limitations as insignificant extra-solution data gathering or transmitting and provided evidence as required by MPEP § 2106.05(d) by citing limitations that the courts have identified to be well-understood, routine, or conventional as described in MPEP § 2106.05(d)(II). The rejection clearly addresses all additional elements and show how these elements individually or in combination do not amount to significantly more than the abstract idea. Further, claim that recites limitations that are novelty does not mean those limitations cannot be performed in a human mind or recite a mathematical concept. Applicant’s arguments, filed 08/12/2026, with respect to the rejections of the claims under 103 have been fully considered and are not persuasive. Applicant argues (pages 7-8) The combination of Ceballos, Vepakomma, and Qiao fail to render obvious independent Claims 1 and 22 because the combination fails to describe or suggest a method of operating a master node in a vertical federated learning, vFL, system for training a split neural network, the vFL system including a plurality of workers, the method comprising "in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received" and "calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs." Ceballos and Vepakomma only describe a standard split neural network. Moreover, Ceballos recognizes the problem associated with the cost of clients dropping without providing any suggestion of any type of solution. (See Ceballos, p. 6). The Office Action also concedes that Ceballos and Vepakomma fail to describe or suggest the feature of "in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received." (See Office Action, p. 32). However, the Office Action asserts that Qiao discloses this feature. Qiao fails to bridge the substantial gap left by Ceballos and Vepakomma because Qiao also fails to describe or suggest the feature of "in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received." The Office Action asserts that Qiao discloses this feature in FIG. 3 and paragraph [0041]. (See Office Action, p. 32). However, Qiao discloses a method for imputing values in a recurrent neural network that analyzes a time series data. Notably, a recurrent neural network and a split neural network are two completely different types of neural networks that operate in two different ways. A recurrent neural network uses an arranged series of nodes where each node has memory of previous values. Thus, imputing values in a recurrent neural network is well known. In contrast, in a split neural network, the nodes are distributed. The master node processes data from worker nodes without access to the raw data. Each worker node processes different raw data. As a result, if a node fails to participate and data is missing for a node, there currently is no way to process the missing data. This problem is recognized by Ceballos without providing any solution. Thus, Qiao cannot be meaningfully combined with Ceballos and Vepakomma because the Office Action does not provide any explanation as to how imputing a value within the context of a recurrent neural network could be applied to split neural network. In response The examiner respectfully disagrees. Ceballos in abstract, pages 1-5, Figs. 1, 3 and 5-7, teaches a Vertical Split neural network architecture for training comprises a server and a plurality of clients. At each iteration, each client uses a local/sub model to compute an output and send it to the server, and the global/shared model is trained with the outputs of all local/sub models. Ceballos also teaches, at each iteration, the system detects that one or more of those outputs are missing (clients dropping). Vepakomma on pages 2-3, fig. 2, teaches at each iteration, the outputs from the clients at the cut-layer are sent to a server. It can be seen that the combination of Ceballos and Vepakomma teaches at each iteration, the outputs from the clients at the cut-layer are sent to a server for training the model, and the server detects one or more of those outputs from the clients at the cut-layer that are missing (not received). However, the combination of Ceballos and Vepakomma does not teach “in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received”. Qiao is added to fill in that missing element which is “in response to determining that one or more outputs from the clients are missing, generating imputed values of the missing outputs”. Qiao is not intended to teach the Split neural network, that limitation has already been taught by Ceballos. Qiao in Fig. 3, abstract and paragraph 0041, teaches “imputation is utilized by the computing device to impute any of the current missing values … imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”. Since the combination of Ceballos and Vepakomma teaches at each iteration, one or more output data from the clients at the cut-layer are missing, and Qiao teaches generating the imputed values for the missing data, therefore, the combination of Ceballos, Vepakomma and Qiao teaches the claim limitation “in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received”. Applicant argues (page 8) Furthermore, even if Qiao were to be permissibly combined with Ceballos and Vepakomma, the combination would also fail to describe or suggest "calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs." Qiao only discloses imputing missing values in a time series of data within a recurring neural network. Notably, a recurring neural network does not use gradients generated by a master node. Thus, Qiao fails to calculate any gradients. Moreover, Ceballos and Vepakomma do not provide a solution to the problem of missing data from nodes in a split neural network. Thus, Ceballos and Vepakomma also fail to describe or suggest calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs. Therefore, the Applicant respectfully submits that the combination of Ceballos, Vepkomma, and Qiao fail to render obvious independent Claims 1 and 22. In response The examiner respectfully disagrees. In response to applicant's arguments against the references individually, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413,208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). The Applicant points at each individual reference and argues that the reference does not teach the claim limitations, while the combination of references does read on the claim limitations. The limitation of “calculating gradients for neurons in the cut-layer based on the received layer outputs” is taught by Vepakomma as stated in the Office Action mailed 05/13/2026. Vepakomma on pages 2-3 and Fig. 2 teaches the outputs at the cut layer from the clients are sent to the server, and the gradients are calculated based on the received outputs. However, the combination of Ceballos and Vepakomma does not teach “calculating gradients based on the imputed layer outputs”. Qiao is added to fill in that missing element. Qiao, as mentioned above, teaches the process of imputing the missing data. Qiao, in paragraph 0041, further teaches the gradient is computed based on the imputed data. Therefore, the combination of Ceballos, Vepakomma and Qiao teaches the claim limitation “calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs”. Claim Rejections - 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-19 and 22 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim 1 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites a method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “determining whether layer outputs for the cut-layer were not received from one of the plurality of workers”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “determining” in the context of this claim encompasses the user determines if data from each of the entities are received. The limitation of “in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “generating” in the context of this claim encompasses the user assigns a value to any missing data that is not received from each of the entities. The limitation of “calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs” recites a mathematical concept. The limitation of “splitting the gradients into groups associated with respective ones of the plurality of workers”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “splitting … into groups” in the context of this claim encompasses the user divides data into groups based on the data values. Step 2A (prong 2): This judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of “a master node”, “a vertical federated learning, vFL, system”, “a split neural network” and “a plurality of workers”. The additional elements are recited at a high-level of generality (i.e., as a generic device performing the generic computer functions) such that they amount no more than mere instructions to apply the exception using the generic computer components (MPEP 2106.05(f)). Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The additional elements of “receiving layer outputs for a sample period from one or more of the plurality of workers for a cut-layer at which the neural network is split between the plurality of workers and the master node” and “transmitting the groups of gradients to respective ones of the plurality of workers” amount to insignificant extra-solution activities of data gathering and transmitting which do not amount to significantly more than the abstract idea (MPEP 2106.05(g)). Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of “a master node”, “a vertical federated learning, vFL, system”, “a split neural network” and “a plurality of workers” to perform the steps amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements of “receiving layer outputs for a sample period from one or more of the plurality of workers for a cut-layer at which the neural network is split between the plurality of workers and the master node” and “transmitting the groups of gradients to respective ones of the plurality of workers” are recited at a high level of generality and amount to insignificant extra-solution activities related to mere data gathering and transmitting (MPEP 2106.05(g)). The courts have found limitations directed to receiving and transmitting information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”). Claim 2 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “determining whether layer outputs for the cut-layer were not received from the one of the plurality of workers from which layer outputs were not received for more than a threshold number of sample intervals”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “determining” in the context of this claim encompasses the user compares two set of data values. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “in response to determining …, re-shaping the cut-layer of the split neural network to exclude neurons associated with the one of the plurality of workers from which layer outputs were not received”. This additional element is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using generic computer component. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)). Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “in response to determining …, re-shaping the cut-layer of the split neural network to exclude neurons associated with the one of the plurality of workers from which layer outputs were not received” amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Claim 3 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “determining a new training batch size, cut layer, and/or neuron count for the split neural network based on the re-shaped cut-layer”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “determining” in the context of this claim encompasses the user determines a training data set for a model. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Indeed, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Claim 4 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “determining whether an accuracy of the split neural network is increasing or decreasing after a training round”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “determining” in the context of this claim encompasses the user observes model performance. The limitation of “in response to determining that the accuracy of the neural network is increasing, reducing a network footprint of the split neural network”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “reducing” in the context of this claim encompasses the user adjusts the training data/parameters of the model based on the performance of the model. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Indeed, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Claim 5 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “reducing the network footprint of the neural network comprises performing at least one of: reducing a training batch size of the neural network, reducing a number of neurons in a cut-layer of the split neural network, and increasing a drop-out rate for neurons in the cut-layer of the split neural network”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “reducing the network footprint of the neural network comprises performing at least one of: reducing a training batch size of the neural network, reducing a number of neurons in a cut-layer of the split neural network, and increasing a drop-out rate for neurons in the cut-layer of the split neural network” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 6 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “informing the plurality of workers of the change to the network footprint”. This limitation amounts to insignificant extra-solution activities of data outputting/presenting (MPEP 2106.05(g)). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the claim recites the additional element of “informing the plurality of workers of the change to the network footprint”. The courts have similarly found limitations directed to [displaying, presenting, outputting] a result, recited at a high level of generality, to be well-understood, routine, and conventional. See (MPEP 2106.05(d)(II), "presenting offers and gathering statistics.") Claim 7 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “in response to determining that the accuracy of the neural network is decreasing, performing at least one of: increasing a training batch size of the neural network …”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “increasing” in the context of this claim encompasses the user adjusts the training data/parameters of the model based on the performance of the model. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “in response to determining that the accuracy of the neural network is decreasing, performing at least one of: increasing a training batch size of the neural network, increasing a number of neurons in a cut-layer of the split neural network, and reducing a drop-out rate for neurons in the cut-layer of the split neural network”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “in response to determining that the accuracy of the neural network is decreasing, performing at least one of: increasing a training batch size of the neural network, increasing a number of neurons in a cut-layer of the split neural network, and reducing a drop-out rate for neurons in the cut-layer of the split neural network” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 8 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “generating a moving average of an accuracy score associated with the split neural network” recites a mathematical concept. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “determining whether the accuracy of the split neural network is increasing or decreasing comprises generating a moving average of an accuracy score associated with the split neural network”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “determining whether the accuracy of the split neural network is increasing or decreasing comprises generating a moving average of an accuracy score associated with the split neural network” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 9 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “the accuracy score comprises an F1- score”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “the accuracy score comprises an F1- score” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 10 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “generating synthetic values of the layer outputs that were not received using a generative model based on previously received values of the layer outputs that were not received”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “generating synthetic values of the layer outputs that were not received using a generative model based on previously received values of the layer outputs that were not received” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 11 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. In particular, the claim recites the additional element of “the generative model takes into account previously received values of layer outputs other than the layer outputs that were not received in addition to the layer outputs that were not received”. This limitation is recited at a high-level of generality (i.e., as a generic device performing the generic computer function) such that it amounts to no more than mere instructions to apply the exception using the generic computer components (MPEP 2106.05(f)). Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The additional element of “the generative model takes into account previously received values of layer outputs other than the layer outputs that were not received in addition to the layer outputs that were not received” amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Claim 12 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “the generative model comprises a multivariate timeseries model”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “the generative model comprises a multivariate timeseries model” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 13 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “smoothing the imputed values of the layer outputs that were not received using an exponential smoothing algorithm based on a previously used layer output value”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “smoothing the imputed values of the layer outputs that were not received using an exponential smoothing algorithm based on a previously used layer output value” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 14 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “smoothing is performed according to the operation: At'<- wAt + (1-w)At-1 where At' is the layer output to be used at sample interval t, At is the imputed layer output at sample interval t, At-1 is the layer output previously used at sample interval t-1, and w is a smoothing weight with 0 <w < 1” recites a mathematical concept. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Indeed, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Claim 15 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites a method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “determining that new layer outputs are being received from the one of the plurality of workers from which layer outputs were previously not received”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “determining” in the context of this claim encompasses the user determines if data from one of the entities is received. The limitation of “smoothing the new layer outputs …” recites a mathematical concept. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “smoothing the new layer outputs based on a previously used imputed layer output”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “smoothing the new layer outputs based on a previously used imputed layer output” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 16 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “smoothing is performed according to the operation: At'<- wAt-1 + (1-w) Rt where At' is the layer output to be used at sample interval t, Rt is the actual layer output at sample interval t, At-1 is the layer output previously used at sample interval t-1, and w is a smoothing weight with 0 <w < 1 recites a mathematical concept. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Indeed, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Claim 17 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “the plurality of workers comprise nodes in a wireless communication network”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “the plurality of workers comprise nodes in a wireless communication network” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 18 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “identifying a trusted neighbor node of the one of the plurality of workers”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “identifying” in the context of this claim encompasses the user identifies a specific entity. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim recites the additional element of “obtaining a version of the layer outputs that were not received from the one of the plurality of workers from the trusted neighbor node”. This limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not integrate into a practical application (see MPEP 2106.05(h)). Step 2B: The claim does not include additional element that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “obtaining a version of the layer outputs that were not received from the one of the plurality of workers from the trusted neighbor node” amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitation that amounts to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the judicial exception (see MPEP 2106.05(h)). Claim 19 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites the method which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “combining the version of the layer outputs that were obtained from the trusted neighbor node with a previously imputed set of layer outputs for the one of the plurality of workers”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “combining” in the context of this claim encompasses the user combines the two set of data. Step 2A (prong 2): This judicial exception is not integrated into a practical application. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Indeed, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the claim does not recite any additional element beside the limitation that can be performed in a human mind. The claim is not patent eligible. Claim 22 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: The claim recites a master node which falls within at least one of the four statutory categories of patent eligible subject matter. Step 2: Step 2A (prong 1): The limitation of “determining whether layer outputs for the cut-layer were not received from one of the one or more workers”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “determining” in the context of this claim encompasses the user determines if data from each of the entities are received. The limitation of “in response to determining that layer outputs for the cut-layer were not received from one of the one or more workers, generating imputed values of the layer outputs that were not received”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “generating” in the context of this claim encompasses the user assigns a value to any missing data that is not received from each of the entities. The limitation of “calculating gradients for neurons in the cut-layer based on the received layer outputs and the imputed layer outputs” recites a mathematical concept. The limitation of “splitting the gradients into groups associated with respective ones of the one or more workers”, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, “splitting … into groups” in the context of this claim encompasses the user divides data into groups based on the data values. Step 2A (prong 2): This judicial exception is not integrated into a practical application. In particular, the claim recites the additional elements of “a master node”, “a vertical federated learning, vFL, system”, “processing circuitry”, “memory “, “a neural network” and “one or more workers”. The additional elements are recited at a high-level of generality (i.e., as a generic device performing the generic computer functions) such that they amount no more than mere instructions to apply the exception using the generic computer components (MPEP 2106.05(f)). Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The additional elements of “receiving layer outputs for a sample period from one or more of the workers for a cut-layer at which the neural network is split between the one or more workers and the master node” and “transmitting the groups of gradients to the respective ones of the one or more workers” amount to insignificant extra-solution activities of data gathering and transmitting which do not amount to significantly more than the abstract idea (MPEP 2106.05(g)). Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of “a master node”, “a vertical federated learning, vFL, system”, “processing circuitry”, “memory “, “a neural network” and “one or more workers” to perform the steps amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements of “receiving layer outputs for a sample period from one or more of the workers for a cut-layer at which the neural network is split between the one or more workers and the master node” and “transmitting the groups of gradients to the respective ones of the one or more workers” are recited at a high level of generality and amount to insignificant extra-solution activities related to mere data gathering and transmitting (MPEP 2106.05(g)). The courts have found limitations directed to receiving and transmitting information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”). 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 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 and 22 are rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. (SplitNN-driven Vertical Partitioning - Applicant provided NPL from the IDS) in view of Vepakomma et al. (Split learning for health: Distributed deep learning without sharing raw patient data) and further in view of Qiao et al. (US Pub. 2022/0197977). As per claim 1, Ceballos teaches a method of operating a master node in a vertical federated learning, vFL, system for training a split neural network, the vFL system including a plurality of workers [abstract, “In this work, we introduce SplitNN-driven Vertical Partitioning, a configuration of a distributed deep learning method called SplitNN to facilitate learning from vertically distributed features”; pages 2-3, Fig. 1, section 3 disclose a Vertical SplitNN architecture comprises a server and a plurality of clients; Examiner interprets the server node as a master node and the clients as the workers”; Vertical SplitNN architecture: Each client computes a fixed portion of the computation graph and passes it to the server], the method comprising: receiving layer outputs for a sample period from one or more of the plurality of workers [pages 2-3, Fig. 1, sections 2-3, “we propose to train partial neural networks (NN) on each client and then aggregate all of their outputs … Vertical SplitNN architecture: Each client computes a fixed portion of the computation graph and passes it to the server … an approach called split learning in which a smaller fraction of the model is present on each client-network and just the output of these models is shared with the server in every iteration”]; determining whether layer outputs for the cut-layer were not received from one of the plurality of workers [pages 5-7, Fig. 3, table 4, section 4.3, “In Table 4, we present the results of dropping some of the clients randomly both during training and testing. The drop during the training means that the model is trained with the outputs of all models but on each iteration one or more of those outputs is missing. On the other hand, dropping during testing means that the model was trained with the outputs of all the models but for the prediction on the test set the output of some of the models from the client side is missing … in Figure 3 we can see that dropping more than two clients, in a four client setting”]; calculating gradients for neurons based on the received layer outputs [page 3, 4th-5th paragraphs, “In our method each partial clients-network encodes its data into a different space and then transmits it to train a shared deep servers-network … Gradients can be backpropagated over each layer to generate gradients of previous layers and to update the current layer”]; splitting the gradients into groups associated with respective ones of the plurality of workers [page 3, last paragraph to page 4, 1st paragraph, “When extending this for multiple concurrent clients as in the SplitNN-driven vertical partitioning case, the backpropagated error will be split and each client-network will be passed the corresponding gradients. Let's take concatenation as an example. The full forward pass will be a concatenation of the forward passes coming from each client-network {F1 * F2 * ... Fk}, k E {1, ..., K} where K is the maximum number of clients. In the same way, the gradients on the concatenation layer will be {L1 * £ 2 * ... Lk}- Thus, we only need to split this gradients before passing them to the upstream clients”]; and transmitting the groups of gradients to the respective ones of the plurality of workers [page 3, last paragraph to page 4, 1st paragraph, “When extending this for multiple concurrent clients as in the SplitNN-driven vertical partitioning case, the backpropagated error will be split and each client-network will be passed the corresponding gradients. Let's take concatenation as an example. The full forward pass will be a concatenation of the forward passes coming from each client-network {F1 * F2 * ... Fk}, k E {1, ..., K} where K is the maximum number of clients. In the same way, the gradients on the concatenation layer will be {L1 * £ 2 * ... Lk}- Thus, we only need to split this gradients before passing them to the upstream clients”]. Ceballos does not explicitly teach receiving layer outputs for a sample period from one or more of the plurality of workers for a cut-layer at which the neural network is split between the plurality of workers and the master node (emphasis added); in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received (emphasis added); calculating gradients for neurons in the cut-layer based on the imputed layer outputs; Vepakomma teaches receiving layer outputs for a sample period from one or more of the plurality of workers for a cut-layer at which the neural network is split between the plurality of workers and the master node [pages 2-3, Fig. 2, section 2, “each client, (for example, radiology center) trains a partial deep network up to a specific layer known as the cut layer. The outputs at the cut layer are sent to a server which completes the rest of the training without looking at raw data (radiology images) from clients. This completes a round of forward propagation without sharing raw data. The gradients are now back propagated at the server from its last layer until the cut layer. The gradients at the cut layer (and only these gradients) are sent back to radiology client centers”; Examiner interprets the clients as the workers and the server as the master node]; calculating gradients for neurons in the cut-layer [pages 2-3, Fig. 2, section 2, “each client, (for example, radiology center) trains a partial deep network up to a specific layer known as the cut layer. The outputs at the cut layer are sent to a server which completes the rest of the training without looking at raw data (radiology images) from clients. This completes a round of forward propagation without sharing raw data. The gradients are now back propagated at the server from its last layer until the cut layer. The gradients at the cut layer (and only these gradients) are sent back to radiology client centers”]; It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the process of receiving layer outputs from one or more of the clients for a cut-layer, and calculating gradients for neurons in the cut-layer of Vepakomma. Doing so would help transmitting the gradients at the cut layer to the clients for further processing (Vepakomma, pages 2-3, section 2). Ceballos and Vepakomma do not teach in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received (emphasis added); calculating gradients based on the imputed layer outputs; Qiao teaches in response to determining that layer outputs for the cut-layer were not received from one of the plurality of workers, generating imputed values of the layer outputs that were not received [abstract, “imputation is utilized by the computing device to impute any of the current missing values”; Fig. 3, paragraph 0041, “train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values … The forward imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”; Since Ceballos in pages 5-7, Fig. 3, table 4, section 4.3, teaches the data from one of more outputs for the cut-layer are missing, and Qiao teaches imputation is used to impute the missing data, therefore, the combination of Ceballos and Qiao teaches the above claim limitation]; calculating gradients based on the imputed layer outputs [paragraph 0041, “At block 350, train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values. The training is performed such that the model takes in the input feature vector, which comprises of the multivariate time series data, the triplet encoding for missing values, and the time interval values, and performs a feedforward pass on the recurrent neural network to generate a prediction. The predicted values from the neural network and the ground truth values are used to compute gradients”; It can be seen that the imputation is utilized to generate the prediction which is used to compute the gradient, therefore, the gradient is calculated based in part on the imputation]; It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the process of generating imputed values of the layer outputs that were not received, and calculating gradients based on the imputed layer outputs of Qiao. Doing so would help passing the corresponding gradients to each client network for further process (Ceballos, page 3, last paragraph). As per claim 22, Ceballos teaches a master node of a vertical federated learning, vFL, system [abstract, “In this work, we introduce SplitNN-driven Vertical Partitioning, a configuration of a distributed deep learning method called SplitNN to facilitate learning from vertically distributed features”; pages 2-3, Fig. 1, section 3 disclose a Vertical SplitNN architecture comprises a server and a plurality of clients; Examiner interprets the server node as a master node and the clients as the workers”; Vertical SplitNN architecture: Each client computes a fixed portion of the computation graph and passes it to the server], comprising: receiving layer outputs for a sample period from one or more workers in the vFL [pages 2-3, Fig. 1, sections 2-3, “we propose to train partial neural networks (NN) on each client and then aggregate all of their outputs … Vertical SplitNN architecture: Each client computes a fixed portion of the computation graph and passes it to the server … an approach called split learning in which a smaller fraction of the model is present on each client-network and just the output of these models is shared with the server in every iteration”]; determining whether layer outputs for the cut-layer were not received from one of the one or more workers [pages 5-7, Fig. 3, table 4, section 4.3, “In Table 4, we present the results of dropping some of the clients randomly both during training and testing. The drop during the training means that the model is trained with the outputs of all models but on each iteration one or more of those outputs is missing. On the other hand, dropping during testing means that the model was trained with the outputs of all the models but for the prediction on the test set the output of some of the models from the client side is missing … in Figure 3 we can see that dropping more than two clients, in a four client setting”]; calculating gradients for neurons based on the received layer outputs [page 3, 4th-5th paragraphs, “In our method each partial clients-network encodes its data into a different space and then transmits it to train a shared deep servers-network … Gradients can be backpropagated over each layer to generate gradients of previous layers and to update the current layer”]; splitting the gradients into groups associated with respective ones of the one or more workers [page 3, last paragraph to page 4, 1st paragraph, “When extending this for multiple concurrent clients as in the SplitNN-driven vertical partitioning case, the backpropagated error will be split and each client-network will be passed the corresponding gradients. Let's take concatenation as an example. The full forward pass will be a concatenation of the forward passes coming from each client-network {F1 * F2 * ... Fk}, k E {1, ..., K} where K is the maximum number of clients. In the same way, the gradients on the concatenation layer will be {L1 * £ 2 * ... Lk}- Thus, we only need to split this gradients before passing them to the upstream clients”]; and transmitting the groups of gradients to the respective ones of the one or more workers [page 3, last paragraph to page 4, 1st paragraph, “When extending this for multiple concurrent clients as in the SplitNN-driven vertical partitioning case, the backpropagated error will be split and each client-network will be passed the corresponding gradients. Let's take concatenation as an example. The full forward pass will be a concatenation of the forward passes coming from each client-network {F1 * F2 * ... Fk}, k E {1, ..., K} where K is the maximum number of clients. In the same way, the gradients on the concatenation layer will be {L1 * £ 2 * ... Lk}- Thus, we only need to split this gradients before passing them to the upstream clients”]. Ceballos does not explicitly teach processing circuitry; and memory coupled with the processing circuitry, wherein the memory includes instructions that when executed by the processing circuitry causes the master node to perform operations comprising: receiving layer outputs for a sample period from one or more of the workers for a cut-layer at which the neural network is split between the one or more workers and the master node (emphasis added); in response to determining that layer outputs for the cut-layer were not received from one of the one or more workers, generating imputed values of the layer outputs that were not received (emphasis added); calculating gradients for neurons in the cut-layer based on the imputed layer outputs; Vepakomma teaches receiving layer outputs for a sample period from one or more of the workers for a cut-layer at which the neural network is split between the one or more workers and the master node [pages 2-3, Fig. 2, section 2, “each client, (for example, radiology center) trains a partial deep network up to a specific layer known as the cut layer. The outputs at the cut layer are sent to a server which completes the rest of the training without looking at raw data (radiology images) from clients. This completes a round of forward propagation without sharing raw data. The gradients are now back propagated at the server from its last layer until the cut layer. The gradients at the cut layer (and only these gradients) are sent back to radiology client centers”; Examiner interprets the clients as the workers and the server as the master node]; calculating gradients for neurons in the cut-layer [pages 2-3, Fig. 2, section 2, “each client, (for example, radiology center) trains a partial deep network up to a specific layer known as the cut layer. The outputs at the cut layer are sent to a server which completes the rest of the training without looking at raw data (radiology images) from clients. This completes a round of forward propagation without sharing raw data. The gradients are now back propagated at the server from its last layer until the cut layer. The gradients at the cut layer (and only these gradients) are sent back to radiology client centers”]; It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the process of receiving layer outputs from one or more of the clients for a cut-layer, and calculating gradients for neurons in the cut-layer of Vepakomma. Doing so would help transmitting the gradients at the cut layer to the clients for further processing (Vepakomma, pages 2-3, section 2). Ceballos and Vepakomma do not teach processing circuitry; and memory coupled with the processing circuitry, wherein the memory includes instructions that when executed by the processing circuitry causes the master node to perform operations comprising: in response to determining that layer outputs for the cut-layer were not received from one of the one or more workers, generating imputed values of the layer outputs that were not received (emphasis added); calculating gradients based on the imputed layer outputs; Qiao teaches processing circuitry [paragraph 0025, “The memory 130 is communicatively coupled to the processor 110”]; and memory coupled with the processing circuitry, wherein the memory includes instructions that when executed by the processing circuitry causes the master node to perform operations comprising [paragraph 0025, “The memory 130 is communicatively coupled to the processor 110”; paragraphs 0091-0092, “These computer readable program instructions may be provided to a processor of a general purpose computer … The computer readable program instructions may also be loaded onto a computer, other programable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer”]: in response to determining that layer outputs for the cut-layer were not received from one of the one or more workers, generating imputed values of the layer outputs that were not received [abstract, “imputation is utilized by the computing device to impute any of the current missing values”; Fig. 3, paragraph 0041, “train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values … The forward imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”; Since Ceballos in pages 5-7, Fig. 3, table 4, section 4.3, teaches the data from one of more outputs for the cut-layer are missing, and Qiao teaches imputation is used to impute the missing data, therefore, the combination of Ceballos and Qiao teaches the above claim limitation]; calculating gradients based on the imputed layer outputs [paragraph 0041, “At block 350, train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values. The training is performed such that the model takes in the input feature vector, which comprises of the multivariate time series data, the triplet encoding for missing values, and the time interval values, and performs a feedforward pass on the recurrent neural network to generate a prediction. The predicted values from the neural network and the ground truth values are used to compute gradients”; It can be seen that the imputation is utilized to generate the prediction which is used to compute the gradient, therefore, the gradient is calculated based in part on the imputation]; It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the process of generating imputed values of the layer outputs that were not received, and calculating gradients based on the imputed layer outputs of Qiao. Doing so would help passing the corresponding gradients to each client network for further process (Ceballos, page 3, last paragraph). Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Fogarty et al. (US Patent 12,125,067). As per claim 2, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Ceballos teaches determining whether layer outputs for the cut-layer were not received from the one of the plurality of workers from which layer outputs were not received for more than a threshold number of sample intervals [page 6, 2nd paragraph, “in Figure 3 we can see that dropping more than two clients, in a four client setting, even affects the convergence of the model and the loss starts to rise by the end of the training, indicating that optimization is drifting from local/global minima. This performance drop however does not arise if a client drops just on a few iterations but is present for most of the training”; It can be seen that the system detects at each iteration, the number of clients dropping, and the number of iterations that the client drops (“a client drops just on a few iterations” (or exceed a number of iteration/threshold number of sample intervals) to determine the performance of the model based on the dropping]; Ceballos, Vepakomma and Qiao does not teach in response to determining that layer outputs for the cut-layer were not received from the one of the plurality of workers from which layer outputs were not received for more than the threshold number of sample intervals, re-shaping the cut-layer of the split neural network to exclude neurons associated with the one of the workers from which layer outputs were not received. Fogarty teaches in response to determining that layer outputs for the cut-layer were not received from the one of the plurality of workers from which layer outputs were not received for more than the threshold number of sample intervals, re-shaping the cut-layer of the split neural network to exclude neurons associated with the one of the workers from which layer outputs were not received [Col. 8, lines 59-67, “identifying each variable in the historical profile data structures that is missing a value for at least one of the multiple historical entities, removing each variable in the historical profile data structures that is missing a value for a number of the multiple historical entities that is greater than a specified minimum entity threshold, and for each of the multiple historical entities that is missing a value for one of the identified variables, imputing an assigned value to the identified variable”; Since Ceballos (as modified) teaches each entity trains a partial network and provides output to the cut layer (Vepakomma, pages 2-3, Fig. 2), and Ceballos on page 6 teaches a process of determining one or more entities with missing data for a number of iterations, while Fogarty teaches a process of removing entities that data is missing more than a threshold, therefore, the combination of Ceballos (as modified) and Fogarty teaches removing one or more entities that have the missing data for an amount of iterations. It can be seen that the cut layer comprising the outputs generated from the entities, so if one or more entities are removed because of the missing data, the cut layer is changed/reshaped, therefore, the combination of Ceballos (as modified) and Fogarty teaches the above claim limitation]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the process of removing one or more entities with missing data of Fogarty. Doing so would help training machine learning systems for automated database element processing and prediction output generation (Fogarty, Col. 1, lines 18-20). As per claim 3, Ceballos, Vepakomma, Qiao and Fogarty teach the method of Claim 2. Fogarty further teaches determining a new training batch size, cut layer, and/or neuron count for the split neural network based on the re-shaped cut-layer [Col. 8, lines 57-67, “the training includes preprocessing the historical profile data structures, and the preprocessing includes identifying each variable in the historical profile data structures that is missing a value for at least one of the multiple historical entities, removing each variable in the historical profile data structures that is missing a value for a number of the multiple historical entities that is greater than a specified minimum entity threshold, and for each of the multiple historical entities that is missing a value for one of the identified variables, imputing an assigned value to the identified variable”; Col. 9, lines 17-19, “removing the determined outlier values from training data for the machine learning mode”; Col. 15, lines 33-37, “The number of neurons can be optimized. At the beginning of training, a network configuration is more likely to have excess nodes. Some of the nodes may be removed from the network during training that would not noticeably affect network performance”; Since Ceballos (as modified) teaches each entity trains a partial network and provides output to the cut layer (Vepakomma, pages 2-3, Fig. 2), and Ceballos on page 6 teaches a process of determining one or more entities with missing data for a number of iterations, while Fogarty teaches a process of removing entities that data is missing more than a threshold, or removing unimportant neuron(s), and it can be seen that the cut layer comprising the outputs generated from the entities, so if one or more entities/neurons are removed because of the missing data, the cut layer is changed/reshaped, and therefore, new (reducing) training batch size is generated based on the re-shaped cut-layer]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the process of determining a new training batch size, cut layer, and/or neuron count for the split neural network based on the re-shaped cut-layer of Fogarty. Doing so would help training machine learning systems for automated database element processing and prediction output generation (Fogarty, Col. 1, lines 18-20). Claims 4-5 and 7 are rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Abdi Taghi Abad et al. (US Patent 10,402,691). As per claim 4, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Ceballos, Vepakomma and Qiao do not teach determining whether an accuracy of the split neural network is increasing or decreasing after a training round; and in response to determining that the accuracy of the neural network is increasing, reducing a network footprint of the split neural network. Abdi Taghi Abad teaches determining whether an accuracy of the split neural network is increasing or decreasing after a training round [Col. 2, lines 3–14, “The one or more training batches may then be used to train the classification model for a number of iterations, e.g., epochs. Based on the training, an accuracy corresponding to each class may be determined. The accuracy may be interpreted broadly and may relate to classification accuracy, precision, recall, F1 score the class, or any combination thereof. Based on the accuracy of each class, the one or more training batches may be adjusted or modified for use in the next training session. For instance, the number of samples associated with the class exhibiting low accuracy may be increased in the training batch to improve the accuracy of that class”; Col. 4, lines 25-32, “the adjustment of the training batch relates to the dynamic increasing or decreasing of the number of samples associated with each class. Because the size of the training batch in this example is constant and set at 100 samples, the number of samples for classes that meet or exceed a desired or a target accuracy level may be decreased so as to increase the number of samples of classes that do not meet the target accuracy”; examiner interprets an accuracy that exceed a desired or a target accuracy level after training at each iteration as an accuracy of the model is increasing, and an accuracy that falls below a desired or a target accuracy level after training at each iteration as an accuracy of the model is decreasing after a training round]; and in response to determining that the accuracy of the neural network is increasing, reducing a network footprint of the split neural network [Col. 4, lines 25-32, “the adjustment of the training batch relates to the dynamic increasing or decreasing of the number of samples associated with each class. Because the size of the training batch in this example is constant and set at 100 samples, the number of samples for classes that meet or exceed a desired or a target accuracy level may be decreased (reducing the training batch) so as to increase the number of samples of classes that do not meet the target accuracy”; Col. 7, lines 2-5, “depending on the accuracies of the classes after training the classification model 410 with the training batch 420, the size of the batch may be adjusted, e.g., dynamically increased or decreased”; It can be seen that if the accuracy of the model exceed the desired level, the size of the training batch is reduced]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include determining whether an accuracy of the split neural network is increasing or decreasing after a training round, and in response to determining that the accuracy of the neural network is increasing, reducing a network footprint of the split neural network of Abdi Taghi Abad. Doing so would help improving the training of the classification model (Abdi Taghi Abad, abstract). As per claim 5, Ceballos, Vepakomma, Qiao and Abdi Taghi Abad teach the method of Claim 4. Abdi Taghi Abad further teaches reducing the network footprint of the neural network comprises performing at least one of: reducing a training batch size of the neural network, reducing a number of neurons in a cut-layer of the split neural network, and increasing a drop-out rate for neurons in the cut-layer of the split neural network [Col. 4, lines 25-32, “the adjustment of the training batch relates to the dynamic increasing or decreasing of the number of samples associated with each class. Because the size of the training batch in this example is constant and set at 100 samples, the number of samples for classes that meet or exceed a desired or a target accuracy level may be decreased (reducing the training batch) so as to increase the number of samples of classes that do not meet the target accuracy”; Col. 7, lines 2-5, “depending on the accuracies of the classes after training the classification model 410 with the training batch 420, the size of the batch may be adjusted, e.g., dynamically increased or decreased”; It can be seen that if the accuracy of the model exceed the desired level, the size of the training batch is reduced]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include reducing the network footprint of the neural network comprises reducing a training batch size the neural network of Abdi Taghi Abad. Doing so would help improving the training of the classification model (Abdi Taghi Abad, abstract). As per claim 7, Ceballos, Vepakomma, Qiao and Abdi Taghi Abad teach the method of Claim 4. Abdi Taghi Abad further teaches in response to determining that the accuracy of the neural network is decreasing, performing at least one of: increasing a training batch size of the neural network, increasing a number of neurons in a cut-layer of the split neural network, and reducing a drop-out rate for neurons in the cut-layer of the split neural network [Col. 4, lines 25-32, “the adjustment of the training batch relates to the dynamic increasing or decreasing of the number of samples associated with each class. Because the size of the training batch in this example is constant and set at 100 samples, the number of samples for classes that meet or exceed a desired or a target accuracy level may be decreased so as to increase the number of samples of classes that do not meet the target accuracy (increasing the training batch size); Col. 7, lines 21-24, “As for classes A and C, classes that have accuracies that fall below the desired accuracy level, samples for classes A and C may be added to quickly and efficiently increase their respective accuracies to or above the desired accuracy”; It can be seen that if the accuracy of the model falls below the desired level, the size of the training batch is increased]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include in response to determining that the accuracy of the neural network is decreasing, increasing a training batch size the neural network of Abdi Taghi Abad. Doing so would help improving the training of the classification model (Abdi Taghi Abad, abstract). Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. in view of Abdi Taghi Abad et al. and further in view of Joshi et al. (US Pub. 2022/0201021). As per claim 6, Ceballos, Vepakomma, Qiao and Abdi Taghi Abad teach the method of Claim 5. Ceballos, Vepakomma, Qiao and Abdi Taghi Abad do not explicitly teach informing the plurality of workers of the change to the network footprint. Joshi teaches informing the plurality of workers of the change to the network footprint [paragraph 0023, “In the illustrated example of FIG. 1, each client cloud 104 is connected to the server cloud 102 via a communication network of the server-side validation manager 106 and each client-side validation manager 110. In some examples, the server-side validation manager 106 sends the client-side validation manager 110 a plan, the plan containing information about one or more tasks and/or updates for the client to process to perform … the plan may include … epoch and/or batch size, a quantity of training dataset iterations”; examiner interprets a process of sending to the client the information containing the batch size as informing the client the network footprint]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include informing the workers of the change to the network footprint of Joshi. Doing so would help creating the privacy preserving models (Joshi, abstract). Claims 8-9 are rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. in view of Abdi Taghi Abad et al. and further in view of Hirose et al. (US Pub. 2010/0100511). As per claim 8, Ceballos, Vepakomma, Qiao and Abdi Taghi Abad teach the method of Claim 4. Abdi Taghi Abad teaches determining whether an accuracy of the split neural network is increasing or decreasing [Col. 2, lines 3–14, “The one or more training batches may then be used to train the classification model for a number of iterations, e.g., epochs. Based on the training, an accuracy corresponding to each class may be determined. The accuracy may be interpreted broadly and may relate to classification accuracy, precision, recall, F1 score the class, or any combination thereof. Based on the accuracy of each class, the one or more training batches may be adjusted or modified for use in the next training session. For instance, the number of samples associated with the class exhibiting low accuracy may be increased in the training batch to improve the accuracy of that class”; Col. 4, lines 25-32, “the adjustment of the training batch relates to the dynamic increasing or decreasing of the number of samples associated with each class. Because the size of the training batch in this example is constant and set at 100 samples, the number of samples for classes that meet or exceed a desired or a target accuracy level may be decreased so as to increase the number of samples of classes that do not meet the target accuracy”]; Ceballos, Vepakomma, Qiao and Abdi Taghi Abad do not teach generating a moving average of an accuracy score associated with the split neural network. Hirose teaches generating a moving average of an accuracy score associated with the split neural network [paragraphs 0003-0004, “an outlier score showing a degree of difference between actual data and data predicted from the learned first statistical model for each data in the data series, and computes the moving average of the outlier scores … Each moving average outlier score is computed based on the learned second statistical model and the moving average of the outlier scores”; examiner interprets the outlier score as the accuracy score which showing the difference between actual and predicted data. Since Ceballos (as modified) teaches a process of determining the accuracy score of the model is increasing or decreasing after each iteration (Abdi Taghi Abad, Col. 2, lines 3-14), and Hirose teaches a process of computing the moving average of the outlier scores, therefore, the combination of Ceballos (as modified) and Hirose teaches the above claim limitation]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include generating a moving average of an accuracy score associated with the split neural network of Hirose. Doing so would help adjusting the training batch to improve the accuracy of the model (Abdi Taghi Abad, Col. 2, lines 9-14). As per claim 9, Ceballos, Vepakomma, Qiao, Abdi Taghi Abad and Hirose teach the method of Claim 8. Ceballos further teaches the accuracy score comprises an F1- score [page 4, section 4, 7th paragraph, “We report accuracy as well as Fl score to account for the class imbalance”]. Claims 10-12 are rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Kim et al. (A Survey of Missing Data Imputation Using Generative Adversarial Networks – Applicant provided NPL from the IDS). As per claim 10, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Qiao teaches generating the imputing values of the layer outputs that were not received comprises generating values of the layer outputs that were not received based on previously received values of the layer outputs that were not received [abstract, “imputation is utilized by the computing device to impute any of the current missing values”; Fig. 3, paragraph 0041, “train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values … The forward imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”]. Ceballos, Vepakomma and Qiao do not teach generating the imputing values of the layer outputs that were not received comprises generating synthetic values of the layer outputs that were not received using a generative model; Kim teaches generating the imputing values of the layer outputs that were not received comprises generating synthetic values of the layer outputs that were not received using a generative model [abstract, “many deep learning models for missing data imputation have been studied. One of the most popular models is Generative Adversarial Networks (GANs), which generate plausible fake data through adversarial training”; page 1, Introduction, last paragraph, “In this paper, we cover a variety of recently proposed GANs for missing data imputation”]; It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include imputing the imputing values using a generative model of Kim. Doing so would help improving the model performance (kim, page 1, Introduction, last paragraph). As per claim 11, Ceballos, Vepakomma, Qiao and Kim teach the method of Claim 10. Qiao further teaches the generative model takes into account previously received values of layer outputs other than the layer outputs that were not received in addition to the layer outputs that were not received [abstract, “imputation is utilized by the computing device to impute any of the current missing values”; Fig. 3, paragraph 0041, “train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values … The forward imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”; Since Qiao teaches imputing is performed using the present value immediately prior to the data that are missing to impute the missing value, while Kim in page 1, abstract and introduction, teaches GAN is used for missing data imputation, therefore, the combination of Ceballos (as modified), Qiao and Kim teaches the above claim limitation]; It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos (as modified) to include the generative model takes into account previously received values of layer outputs of Qiao and Kim. Doing so would help improving the model performance (kim, page 1, Introduction, last paragraph). As per claim 12, Ceballos, Vepakomma, Qiao and Kim teach the method of Claim 10. Qiao further teaches the generative model comprises a multivariate timeseries model [paragraph 0041 discloses multivariate time series data id used to determine the missing values, while Kim in page 1, abstract and introduction, teaches GAN is used for missing data imputation, therefore, the combination of Ceballos (as modified), Qiao and Kim teaches the above claim limitation]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos (as modified) to include the generative model comprises a multivariate timeseries model of Qiao and Kim. Doing so would help improving the model performance (kim, page 1, Introduction, last paragraph). Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Sukatis et al. (Estimate of missing values in air pollution dataset by using various imputation methods). As per claim 13, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Qiao further teaches the imputed values of the layer outputs … based on a previously used layer output value [abstract, “imputation is utilized by the computing device to impute any of the current missing values”; Fig. 3, paragraph 0041, “train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values … The forward imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”]; Ceballos, Vepakomma and Qiao do not teach smoothing the imputed values of the layer outputs that were not received using an exponential smoothing algorithm based on a previously used layer output value. Sukatis teaches smoothing the imputed values of the layer outputs that were not received using an exponential smoothing algorithm based on a previously used layer output value [abstract, “determine the best imputation method to fill in the various gaps of missing values in air pollution dataset. Ten imputation methods such as … Exponential Smoothing were applied to fill in the missing values”]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include smoothing the imputed values using an exponential smoothing algorithm of Sukatis. Doing so would help filling in the simulated of short, long, and combination of short-long gaps of missing data (Sukatis, page 803, conclusion). Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Pinto et al. (US Pub. 2005/0234688). As per claim 15, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Ceballos teaches determining … layer outputs were previously not received [pages 5-7, Fig. 3, table 4, section 4.3, “In Table 4, we present the results of dropping some of the clients randomly both during training and testing. The drop during the training means that the model is trained with the outputs of all models but on each iteration one or more of those outputs is missing. On the other hand, dropping during testing means that the model was trained with the outputs of all the models but for the prediction on the test set the output of some of the models from the client side is missing]. Qiao teaches the imputed values of the layer outputs … based on a previously used layer output value [abstract, “imputation is utilized by the computing device to impute any of the current missing values”; Fig. 3, paragraph 0041, “train, by a computing device, a recurrent neural network with the triplet encodings, the time intervals, and the multivariate time series data to predict values at future time steps, where a forward imputation is utilized by the computing device to predict the random missing values … The forward imputation is performed by carrying forward last present values, i.e., using the present value immediately prior to the data that are missing to impute the missing value”]; Ceballos, Vepakomma and Qiao do not teach determining that new layer outputs are being received from the one of the plurality of workers from which layer outputs were previously not received; and smoothing the new layer outputs based on a previously used imputed layer output. Pinto teaches determining that new layer outputs are being received from the one of the plurality of workers from which layer outputs were previously not received [paragraph 0062, “the model generation platform provides automated and interactive features to transform variables by imputing missing values 120”; examiner interprets the variables with imputed values as the new outputs that were previously not received]; and smoothing the new layer outputs based on a previously used imputed layer output [paragraph 0062, “the model generation platform provides automated and interactive features to transform variables by imputing missing values 120 … smoothing variables with grouping and splines”; examiner interprets the variables with imputed values as the new outputs that were previously not received, and smoothing process is performed on the variables with imputed values]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos (as modified) to include determining that new layer outputs are being received, and smoothing the new layer outputs of Pinto. Doing so would help training the neural network to improve the network performance (Qiao, 0041). Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Milletari et al. (US Patent 11,804,050). As per claim 17, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Ceballos, Vepakomma and Qiao do not explicitly teach the plurality of workers comprise nodes in a wireless communication network. Milletari teaches the plurality of workers comprise nodes in a wireless communication network [Col. 19, lines 30-60, “a client-server network environment. In at least one embodiment, client devices 502 each corresponding to clients … a client device 502 and/or an edge device may be embodied as a personal computer (PC), a laptop computer, a mobile device, a smartphone, a tablet computer, a smart watch, a wearable computer, a personal digital assistant …”]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include the client nodes comprise nodes in a wireless communication network of Milletari. Doing so would help the clients to wirelessly connect to the server via internet (Milletari, Col. 45, lines 62-67) Claim 18 is rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. and further in view of Poulos et al. (Missing Data Imputation for Supervised Learning). As per claim 18, Ceballos, Vepakomma and Qiao teach the method of Claim 1. Ceballos, Vepakomma and Qiao do not explicitly teach identifying a trusted neighbor node of the one of the plurality of workers; and obtaining a version of the layer outputs that were not received from the one of the plurality of workers from the trusted neighbor node. Poulos teaches identifying a trusted neighbor node of the one of the plurality of workers [abstract, “Missing data imputation can help improve the performance of prediction models … k-nearest-neighbors (k-NN) imputation”; page 188, 4th paragraph, “computing the k-NN of an example with missing data and assigning the mode of the k-neighbors to the missing data”]; and obtaining a version of the layer outputs that were not received from the one of the plurality of workers from the trusted neighbor node [page 188, 4th paragraph, “computing the k-NN of an example with missing data and assigning the mode of the k-neighbors to the missing data”]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include obtaining a version of the layer outputs that were not received from the one of the workers from the trusted neighbor node of Poulos. Doing so would help improving the performance of prediction models (Poulos, abstract). Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Ceballos et al. in view of Vepakomma et al. in view of Qiao et al. in view of Poulos et al. and further in view of Welsh et al. (US Pub. 2006/0271300). As per claim 19, Ceballos, Vepakomma, Qiao and Poulos teach the method of Claim 18. Poulos teaches the version of the layer output that were obtained from the trusted neighbor node [page 188, 4th paragraph, “computing the k-NN of an example with missing data and assigning the mode of the k-neighbors to the missing data”]; Ceballos, Vepakomma, Qiao and Poulos do not teach combining the version of the layer outputs that were obtained from the trusted neighbor node with a previously imputed set of layer outputs for the one of the plurality of workers. Welsh teaches combining the version of the layer outputs that were obtained from the trusted neighbor node with a previously imputed set of layer outputs for the one of the plurality of workers [paragraph 0009, “K-nearest neighbors (KNN) and singular value decomposition (SVD) imputation methods have also been used … the KNN imputation method uses local patterns. The K records that are closest to the record with missing data are combined in the estimation”; Since Poulos teaches the mode of each neighbor can be assigned to the missing data, while Welsh teaches combining the K records of the nearest neighbors to perform imputation, therefore, the combination of Ceballos (as modified) and Welsh teaches the above claim limitation]. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the vertical splitNN method of Ceballos to include combining the version of the layer outputs that were obtained from the trusted neighbor node with a previously imputed set of layer outputs of Welsh. Doing so would help improving the missing value estimation (Welsh, 0009). Allowable Subject Matter Claims 14 and 16 are objected to as being dependent upon the rejected base claims, but would be allowable if rewritten in independent form including all of the limitations of the base claims and any intervening claims, and the 101 rejections above were overcome. As per claim 14, The claim was previously indicated as having allowable subject matter, the examiner has performed an updated search, but no prior art references are found to teach or suggest, individually or in combination the limitations: smoothing is performed according to the operation: At'<- wAt + (1-w) At-1 where At' is the layer output to be used at sample interval t, At is the imputed layer output at sample interval t, At-1 is the layer output previously used at sample interval t-1, and w is a smoothing weight with 0 <w < 1. Therefore, the combination of features is considered to be allowable. As per claim 16, The claim was previously indicated as having allowable subject matter, the examiner has performed an updated search, but no prior art references are found to teach or suggest, individually or in combination the limitations: smoothing is performed according to the operation: At'<- wAt-1 + (1-w) Rt where At' is the layer output to be used at sample interval t, Rt is the actual layer output at sample interval t, At-1 is the layer output previously used at sample interval t-1, and w is a smoothing weight with 0 <w < 1. Therefore, the combination of features is considered to be allowable. Prior Art The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure. Guo et al. (US Pub. 2019/0129819) describes a method for imputing corrupted sequential data. Rho et al. (US Pub. 2022/0207300) describes a process of generating missing data as missing data imputation values similar to real data using a generative adversarial network. Conclusion THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any 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 TRI T NGUYEN whose telephone number is 571-272-0103. The examiner can normally be reached M-F, 8 AM-5 PM, (CT). 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, OMAR FERNANDEZ can be reached at 571-272-2589. 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. /TRI T NGUYEN/Examiner, Art Unit 2128 /OMAR F FERNANDEZ RIVAS/Supervisory Patent Examiner, Art Unit 2128
Read full office action

Prosecution Timeline

Aug 11, 2023
Application Filed
May 13, 2026
Non-Final Rejection mailed — §101, §103
Aug 12, 2026
Response Filed
Nov 12, 2026
Final Rejection mailed — §101, §103 (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12783477
PRUNING A DATABASE USING GENETIC ALGORITHM
4y 5m to grant Granted Nov 03, 2026
Patent 12769992
METHOD FOR DESIGNING SCALABLE AND ENERGY-EFFICIENT ANALOG NEUROMORPHIC PROCESSORS
6y 3m to grant Granted Oct 20, 2026
Patent 12749002
ANALOG SYSTEM USING EQUILIBRIUM PROPAGATION FOR LEARNING
6y 3m to grant Granted Sep 29, 2026
Patent 12725018
NEURAL NETWORK ACCELERATOR
6y 0m to grant Granted Sep 01, 2026
Patent 12645964
QUANTUM CONTROL SYSTEM, QUANTUM CONTROL PROCESSOR, AND QUANTUM INSTRUCTION SET EXECUTION METHOD
4y 4m to grant Granted Jun 02, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

3-4
Expected OA Rounds
67%
Grant Probability
82%
With Interview (+14.5%)
3y 12m (~8m remaining)
Median Time to Grant
Moderate
PTA Risk
Based on 203 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month