DETAILED ACTION
This Office Action is sent in response to Applicant’s Communication received 4/20/2026 for application number 17/971,488.
Claims 1, 3-10, 12-22 are pending.
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 .
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 1, 7-8, 10, 16-17, and 19-22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Akdeniz et al. (US 2023/0068386 A1) in view of Dimitriadis et al. (US 2022/0036178 A1).
In reference to claim 1, Akdeniz discloses a data processing method (para. 0102), performed by a central node device in a distributed system (performed by central server in system with the central server and client devices, para. 0127-30), comprising: acquiring model training information transmitted by each of at least two edge node devices of the distributed system (central server receives model updates, para. 0150), the model training information being transmitted in a form of plaintext (encryption, fig. 22 and para. 0446, is optional, compare fig. 22 with figs. 20 and 13; in embodiments without encryption, the updates would be sent in plaintext), and being obtained by the edge node device by training sub-models through differential privacy (training info derived by clients using differential privacy techniques, para. 0148-50; 0435-39); acquiring, based on the model training information transmitted by each of the at least two edge node devices, the sub-models trained by each of the at least two edge node devices (because the model updates are partial gradients for the global model that the client has locally calculated via training, para. 0150, and “the updated global model is effectively a weighted sum of the updated local models,” para. 0190, the server receiving the model updates causes the server to acquire the sub-models from the clients); and performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model, the target model ensemble policy being a model ensemble policy other than a cryptography-based security model fusion policy (client models are aggregated through policy like averaging weight values, para. 0127-28), wherein the target model ensemble policy comprises a first model ensemble policy, the performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model comprises: acquiring, based on the first model ensemble policy, ensemble weights of the sub-models trained by the at least two edge node devices, the ensemble weights being used for indicating impact of output values of the sub-models on an output value of the global model (model updates include weights, para. 0127-28, 0150) … acquiring at least one sub-model from the sub-models trained by the at least two edge node devices, to generate at least one ensemble model set, the ensemble model set being a set of sub-models for ensembling into a global model; and performing weighted averaging on the sub-models in the at least one ensemble model set based on the ensemble weights to obtain the at least one global model (weighted averaging of weights is used to obtain global model, para. 0150).
However, Akdeniz does not explicitly teach the ensemble weights indicating impact of output values of the sub-models on an output value of the global model and the ensemble weights are positively correlated with a weight impact parameter of each of the at least two edge node devices, wherein the weight impact parameter includes at least one of: reliability corresponding to the edge node device and a data amount of a training data set in the edge node device.
Dimitriadis teaches the ensemble weights indicating impact of output values of the sub-models on an output value of the global model and the ensemble weights are positively correlated with a weight impact parameter of each of the at least two edge node devices, wherein the weight impact parameter includes at least one of: reliability corresponding to the edge node device and a data amount of a training data set in the edge node device (local model gradients are combined for global model using a weight factor based on gradient quality, which is reliability, para. 0024-34, figs. 2-3; the Examiner also notes Bai, US 2022/0237507 A1 cited in conclusion below, which teaches an amount of training data).
It would have been obvious to one of ordinary skill in art, having the teachings of Akdeniz and Dimitriadis before the earliest effective filing date, to modify the weights as disclosed by Akdeniz to include the reliability weight as taught by Dimitriadis.
One of ordinary skill in the art would have been motivated to modify the weights of Akdeniz to include the reliability weight of Dimitriadis because it helps better train in federated learning environments (Dimitriadis, para. 0014).
In reference to claim 7, Akdeniz discloses the method according to claim 1, wherein a same differential privacy algorithm is used during the training performed by the at least two edge node devices on the respective sub-models; or different differential privacy algorithms are used during the training performed by the at least two edge node devices on the respective sub-models (same algorithm is used, para. 0429-39).
In reference to claim 8, Akdeniz discloses the method according to claim 1, wherein the at least two first training data sets stored in the at least two edge node devices conform to horizontal federated learning (HFL) data distribution (Akdeniz uses a horizontal federated learning architecture, see architecture in fig. 12 with central server in which training data on client devices is kept secret from a central server; see definition in Applicant’s specification, para. 0004; also see Wen et al., A survey on federated learning: challenges and applications, at pages 515-516, attached NPL).
In reference to claim 10, this claim is directed to a device associated with the method claimed in claim 1 and is therefore rejected under a similar rationale.
In reference to claim 11, this claim is directed to a device associated with the method claimed in claim 2 and is therefore rejected under a similar rationale.
In reference to claim 16, this claim is directed to a device associated with the method claimed in claim 7 and is therefore rejected under a similar rationale.
In reference to claim 17, this claim is directed to a device associated with the method claimed in claim 8 and is therefore rejected under a similar rationale.
In reference to claim 19, this claim is directed to a non-transitory computer-readable medium associated with the method claimed in claim 1 and is therefore rejected under a similar rationale.
In reference to claim 20, this claim is directed to a device associated with the method claimed in claim 8 and is therefore rejected under a similar rationale.
In reference to claim 21, Akdeniz discloses the method of claim 1, wherein the sub-models are trained by the edge device using a respective training data set to generate an intermediate model gradient and random noise is added to the intermediate model gradient through the differential privacy mechanism (using differential privacy, noise can be injected after calculating update, i.e. intermediate model gradient, para. 0429-0445).
In reference to claim 22, Akdeniz discloses the method of claim 1, wherein the sub-models are trained by the edge device by adding random noise to a respective training data set through the differential privacy mechanism (using differential privacy, noise can be injected into raw data, which is training data, before calculating update, para. 0429-0445).
Claim(s) 3-5 and 12-14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Akdeniz et al. (US 2023/0068386 A1) in view of Dimitriadis et al. (US 2022/0036178 A1) as applied to claims 1 and 10 above, and in further view of Joshi et al. (US 2022/0201021 A1).
In reference to claim 3, Akdeniz and Dimitriadis do not explicitly teach the method according to claim 1, wherein the target model ensemble policy comprises a second model ensemble policy, the central node device comprises a second training data set, the second training data set being a data set stored in the central node device, and comprising feature data and label data; and the performing, based on a target model ensemble policy, model ensemble on the sub- models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a first initial global model based on the second model ensemble policy; inputting the feature data in the second training data set to the sub-models trained by the at least two edge node devices, to obtain at least two pieces of first output data; inputting the first output data to the first initial global model; and updating a model parameter in the first initial global model based on the label data in the second training data set and an output result of the first initial global model, to obtain the global model.
Joshi teaches the method according to claim 1, wherein the target model ensemble policy comprises a second model ensemble policy, the central node device comprises a second training data set, the second training data set being a data set stored in the central node device, and comprising feature data and label data (server stores ground truth data used for training and verification, para. 0026-28, 0038, the data including the known legitimate inputs and outputs, para. 0023); and the performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a first initial global model based on the second model ensemble policy (initial global model is trained, para. 0027); inputting the feature data in the second training data set to the sub-models trained by the at least two edge node devices, to obtain at least two pieces of first output data; inputting the first output data to the first initial global model; and updating a model parameter in the first initial global model based on the label data in the second training data set and an output result of the first initial global model, to obtain the global model (local models can be verified using the ground truth data, para. 0030-39, 0063-65, and in response to the output, the global model can be updated based on the output verification data, para. 0063-65).
It would have been obvious to one of ordinary skill in art, having the teachings of Akdeniz, Dimitriadis, and Joshi before the earliest effective filing date, to modify the federated learning aggregation as disclosed by Akdeniz to include the training data as taught by Joshi.
One of ordinary skill in the art would have been motivated to modify the federated learning aggregation of Akdeniz to include the training data of Joshi because it can help prevent poisoning of the ML model during training (Joshi, para. 0015).
In reference to claim 4, Akdeniz and Dimitriadis do not explicitly teach the method according to claim 1, wherein the target model ensemble policy comprises a third model ensemble policy, the central node device comprises a second training data set, the second training data set being a data set stored in the central node device, and comprising feature data and label data; and the performing, based on a target model ensemble policy, model ensemble on the sub- models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a second initial global model based on the third model ensemble policy; inputting the feature data in the second training data set to the sub-models trained by the at least two edge node devices, to obtain at least two pieces of first output data; inputting the first output data and the feature data in the second training data set to the second initial global model to obtain second output data; and updating a model parameter in the second initial global model based on the second output data and the label data in the second training data set, to obtain the global model.
Joshi teaches the method according to claim 1, wherein the target model ensemble policy comprises a third model ensemble policy, the central node device comprises a second training data set, the second training data set being a data set stored in the central node device, and comprising feature data and label data (server stores ground truth data used for training and verification, para. 0026-28, 0038, the data including the known legitimate inputs and outputs, para. 0023); and the performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a second initial global model based on the third model ensemble policy (initial global model is trained, para. 0027); inputting the feature data in the second training data set to the sub-models trained by the at least two edge node devices, to obtain at least two pieces of first output data (local models can be verified using the ground truth data, para. 0030-39, 0063-65); inputting the first output data and the feature data to the second initial global model to obtain second output data (using verification output and the ground truth data, the global model can be trained and verified, para. 0023-28); and updating a model parameter in the second initial global model based on the second output data and the label data in the second training data set, to obtain the global model (global model is updated based on outputs and ground truth data, para. 0030-39, 0063-65).
It would have been obvious to one of ordinary skill in art, having the teachings of Akdeniz, Dimitriadis, and Joshi before the earliest effective filing date, to modify the federated learning aggregation as disclosed by Akdeniz to include the training data as taught by Joshi.
One of ordinary skill in the art would have been motivated to modify the federated learning aggregation of Akdeniz to include the training data of Joshi because it can help prevent poisoning of the ML model during training (Joshi, para. 0015).
In reference to claim 5, Akdeniz and Dimitriadis do not explicitly teach the method according to claim 1, wherein the target model ensemble policy comprises a fourth model ensemble policy, the central node device comprises a second training data set, the second training data set being a data set stored in the central node device, and comprising feature data and label data; and the performing, based on a target model ensemble policy, model ensemble on the sub- models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a third initial global model based on the fourth model ensemble policy, the third initial global model being a classification model; inputting the feature data in the second training data set to the sub-models trained by the at least two edge node devices, to obtain at least two pieces of first output data; collecting statistics on classification results of the first output data in response to the first output data being classification result data, to obtain a statistical result corresponding to each of the classification results; and updating a model parameter in the third initial global model based on the statistical result and the label data, to obtain the global model
Joshi teaches the method according to claim 1, wherein the target model ensemble policy comprises a fourth model ensemble policy, the central node device comprises a second training data set, the second training data set being a data set stored in the central node device, and comprising feature data and label data (server stores ground truth data used for training and verification, para. 0026-28, 0038, the data including the known legitimate inputs and labels, para. 0023); and the performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a third initial global model based on the fourth model ensemble policy, the third initial global model being a classification model (models trained to classify data as malicious, para. 0031, 42); inputting the feature data in the second training data set to the sub-models trained by the at least two edge node devices, to obtain at least two pieces of first output data (local models can be verified using the ground truth data, para. 0030-39, 0063-65); collecting statistics on classification results of the first output data in response to the first output data being classification result data, to obtain a statistical result corresponding to each of the classification results; and updating a model parameter in the third initial global model based on the statistical result and the label data, to obtain the global model (accuracy metrics are calculated, which are statistics of the classification results, and the global model is updated based if the accuracy metrics are above a threshold, para. 0030-39, 0063-65).
It would have been obvious to one of ordinary skill in art, having the teachings of Akdeniz, Dimitriadis, and Joshi before the earliest effective filing date, to modify the federated learning aggregation as disclosed by Akdeniz to include the training data as taught by Joshi.
One of ordinary skill in the art would have been motivated to modify the federated learning aggregation of Akdeniz to include the training data of Joshi because it can help prevent poisoning of the ML model during training (Joshi, para. 0015).
In reference to claim 12, this claim is directed to a device associated with the method claimed in claim 3 and is therefore rejected under a similar rationale.
In reference to claim 13, this claim is directed to a device associated with the method claimed in claim 4 and is therefore rejected under a similar rationale.
In reference to claim 14, this claim is directed to a device associated with the method claimed in claim 5 and is therefore rejected under a similar rationale.
Claim(s) 6 and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Akdeniz et al. (US 2023/0068386 A1) in view of Dimitriadis et al. (US 2022/0036178 A1) as applied to claims 1 and 10 above, and in further view of Beaufays et al. (US 2022/0270590 A1).
In reference to claim 6, Akdeniz and Dimitriadis do not explicitly teach the method according to claim 1, wherein the target model ensemble policy comprises a fifth model ensemble policy, the performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a functional layer of at least one sub-model from the sub-models corresponding to the edge node devices based on the fifth model ensemble policy, the functional layer being configured to indicate a partial model structure that implements a specified functional operation; and acquiring a model comprising at least two functional layers as the global model in response to the model composed of the at least two functional layers having a complete model structure.
Beaufays teaches the method according to claim 1, wherein the target model ensemble policy comprises a fifth model ensemble policy, the performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model comprises: acquiring a functional layer of at least one sub-model from the sub-models corresponding to the edge node devices based on the fifth model ensemble policy, the functional layer being configured to indicate a partial model structure that implements a specified functional operation; and acquiring a model comprising at least two functional layers as the global model in response to the model composed of the at least two functional layers having a complete model structure (portions of local ML model layers are used to build global ML model layers, para. 0066, and then additional layers are combined with the global ML model layers to produce a model with 2+ layers, para. 0067).
It would have been obvious to one of ordinary skill in art, having the teachings of Akdeniz, Dimitriadis, and Beaufays before the earliest effective filing date, to modify the federated learning aggregation as disclosed by Akdeniz to include the layers as taught by Beaufays.
One of ordinary skill in the art would have been motivated to modify the federated learning aggregation of Akdeniz to include the layers of Beaufays because it can help train more quickly and result in higher accuracy models (Beaufays, para. 0014).
In reference to claim 15, this claim is directed to a device associated with the method claimed in claim 6 and is therefore rejected under a similar rationale.
Claim(s) 9 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Akdeniz et al. (US 2023/0068386 A1) in view of Dimitriadis et al. (US 2022/0036178 A1) as applied to claims 1 and 10 above, and in further view of Verma et al. (US 12,033,047 B2).
In reference to claim 9, Akdeniz and Dimitriadis do not explicitly teach the method according to claim 1, wherein the model structures of the sub-models trained by the at least two edge node devices are different.
Verma teaches the method according to claim 1, wherein the model structures of the sub-models trained by the at least two edge node devices are different (clients can have different model architectures, col. 10, lines 33-41).
It would have been obvious to one of ordinary skill in art, having the teachings of Akdeniz, Dimitriadis, and Verma before the earliest effective filing date, to modify the federated learning aggregation as disclosed by Akdeniz to include the different client architectures as taught by Verma.
One of ordinary skill in the art would have been motivated to modify the federated learning aggregation of Akdeniz to include the different client architectures of Verma because it would allow for federated learning without dependency on homogenous model architectures (Verma, para. 0014).
In reference to claim 18, this claim is directed to a device associated with the method claimed in claim 9 and is therefore rejected under a similar rationale.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Bai (US 20220237507 A1) which teaches weighting local updates based on a number of data points in local training data set, para. 0117.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Andrew T. Chiusano whose telephone number is (571)272-5231. The examiner can normally be reached M-F, 10am-6pm.
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, Tamara Kyle can be reached at 571-272-4241. 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.
/ANDREW T CHIUSANO/Primary Examiner, Art Unit 2144
Prosecution Insights