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 .
This action is in response to communications filed on 04/16/2026.
Claims 9 and 21-22 have been canceled.
Claims 1-8, 10-20, and 23 are pending and have been examined.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 03/24/2026 has been entered.
Priority
Applicant’s claim for the benefit of a prior-filed application under 35 U.S.C. 119(e) or under 35 U.S.C. 120, 121, 365(c), or 386(c) is acknowledged.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 5 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
As per claim 5, there is lack of antecedent basis for “the unvalidated” in line 3.
Response to Arguments
Applicant’s arguments with respect to newly amended features have been considered but are moot in view of new grounds of rejection. See combination in view of Alistarh below.
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.
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.
Claims 1-4, 6, 11-16, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (US 20190042878 A1) in view of Tomsett et al. (US 20200402658 A1) and Alistarh et al. ("Byzantine Stochastic Gradient Descent", arXiv:1803.08917v1, March 23 2018, pages 1-20).
As per independent claim 1, Sheller teaches a method performed by a server node for generating a machine learning (ML) model (e.g. in paragraph 13), the method comprising:
providing an initial version of the ML model to at least a first unvalidated client device (e.g. in paragraphs 21-22, “example aggregator device 110 provides a current state of the machine learning model to each edge device [including] non-trusted edge devices”);
receiving, from the first unvalidated client device, updated model parameters for a first locally trained ML model that was trained using local data available to the first unvalidated client device and starting from the initial version of the ML model (e.g. in paragraph 21-22, “provides a current state of the machine learning model… Each edge device may then perform local training and provide training results to the aggregator device 110 for aggregation… model updates from… non-trusted edge devices”);
obtaining a first logical explanation for the first unvalidated client device (e.g. in paragraph 47, “provides a model update to the example aggregator 110. In some examples, additional information is provided along with the model update such as, for example, an identity of the edge device 130, an indication of how much training data was used to prepare the model update, and/or other parameters identified as part of the model training process… provides model information concerning the private layers of the model (e.g., model updates resulting from training by the private model trainer 320)… may additionally provide model information concerning the public layers of the model”);
deciding, based on a first distance value, whether or not to use the updated model parameters received from the first unvalidated client device to generate the ML model (e.g. in paragraphs 22 and 63, “example aggregator device 110 applies Byzantine Gradient Descent to model updates that originate from non-trusted edge devices. (Block 445). Applying Byzantine Gradient Descent to model updates originating from non-trusted edge devices enables elimination of extreme model updates (which may potentially be malicious)”, i.e. updates that are significantly outside [i.e. “distance”] trusted model updates); if it is decided to use the updated model parameters received from the first unvalidated client device to generate the ML model, then generating the ML model using a set of updated model parameters that includes the updated model parameters received from the first validated client device and the updated model parameters received from the first unvalidated client device (e.g. in paragraph 21, “aggregator device 110 accesses the results provided by the edge devices 130, 137. In some examples, the model updates are aggregated as they arrive at the aggregator device”, i.e. both if not malicious), otherwise generating the ML model using a set of updated model parameters that includes the updated model parameters received from the first validated client device but that does not include the updated model parameters received from the first unvalidated client device (e.g. in paragraphs 22-23 and 63-64, “example aggregator device 110 aggregates model updates from trusted edge devices. (Block 443). That is, if a model update is received from a trusted edge device (e.g., an edge device that implements a trusted execution environment), it is automatically included in the aggregation… elimination of extreme model updates (which may potentially be malicious)… example aggregator device 110 updates a centrally stored model”, i.e. only trusted if malicious); and
outputting the generated ML model (e.g. in paragraphs 22-23 and 63-64, “updates a centrally stored model”),
but does not specifically teach the first logical explanation based on the updated model parameters received from the first unvalidated client device and at least a first set of input values and a first set of corresponding output values for the first unvalidated client device and using the first logical explanation for the unvalidated client device and a second logical explanation that was obtained using updated model parameters received from a first validated client device, obtaining a first distance value indicating a distance between the first logical explanation for the first unvalidated client device and the second logical explanation obtained using the updated model parameters received from the first validated client device.
However, Tomsett teaches a logical explanation(s) based on updated model parameters received from a device and at least a first set of input values and a first set of corresponding output values for the device (e.g. in paragraphs 97 and 105, “explanation generation components 102.sub.1-N can generate output data… the output data can include a set of model parameters, a set of model weights for a model (e.g., a neural network model) associated with the input data 108 and... a set of gradients…associated with model data… e.g., further processed audio, video, textual and/or numerical data; further updated model parameter, model weights and/or gradients”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Sheller to include the teachings of Tomsett because one of ordinary skill in the art would have recognized the benefit of using relevant information,
but does not specifically teach using the first logical explanation for the unvalidated client device and a second logical explanation that was obtained using updated model parameters received from a first validated client device, obtaining a first distance value indicating a distance between the first logical explanation for the first unvalidated client device and the second logical explanation obtained using the updated model parameters received from the first validated client device.
However, Alistarh teaches using a first logical explanation for an unvalidated client device and a second logical explanation that was obtained using updated model parameters received from a first validated client device, obtaining a first distance value indicating a distance between the first logical explanation for the first unvalidated client device and the second logical explanation obtained using the updated model parameters received from the first validated client device (e.g. in pages 2 and 4, “a good (i.e., non-Byzantine) worker machine returns Vfs(xk)… [first logical explanation] should concentrate around [second logical explanation obtained from good/“validated” machines] for each good machine i, up to an additive error… In other words, if [distance>threshold] for some machine i, we can safely declare i is a Byzantine machine”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Alistarh because one of ordinary skill in the art would have recognized the benefit of facilitating Byzantine Gradient Descent to determine malicious devices.
As per claim 2, the rejection of claim 1 is incorporated and the combination further teaches wherein the first distance value is less than a predetermined threshold, and the updated model parameters received from the first unvalidated client device are aggregated with the updated model parameters from the first validated client device to generate the ML model (e.g. Sheller, in paragraphs 22-23 and 63-64, “example aggregator device 110 aggregates model updates from trusted edge devices. (Block 443). That is, if a model update is received from a trusted edge device (e.g., an edge device that implements a trusted execution environment), it is automatically included in the aggregation… elimination of extreme model updates (which may potentially be malicious)… example aggregator device 110 updates a centrally stored model”; Alistarh, in pages 2 and 4, “up to an additive error [distance>threshold] … In other words, if [distance>threshold] for some machine i, we can safely declare i is a Byzantine machine”, i.e. less than the threshold is not malicious).
As per claim 3, the rejection of claim 1 is incorporated and the combination further teaches generating a secondary ML model based on the updated model parameters received from the first unvalidated client device (e.g. Sheller, in paragraphs 63-64, “updates a centrally stored model. (Block 450). The updated model then serves as a new model for the next training iteration”).
As per claim 4, the rejection of claim 1 is incorporated and the combination further teaches determining whether the first distance value is larger than a pre-defined distance threshold and removing the first unvalidated client device as a result of determining that the first distance value is larger than the pre-defined distance threshold (e.g. Sheller, in paragraphs 22-23 and 63-64, “elimination of extreme model updates (which may potentially be malicious)”; Alistarh, in pages 2 and 4, “up to an additive error [distance>threshold] … In other words, if [distance>threshold] for some machine i, we can safely declare i is a Byzantine machine”, i.e. malicious).
As per claim 6, the rejection of claim 1 is incorporated and the combination further teaches wherein the ML model is a neural network and the updated model parameters are weights (e.g. Sheller, in paragraph 93, “enable distributed training of a neural network that is robust against potential attack vectors”; Tomsett, in paragraph 97, “a set of model weights for a model”).
Claims 11-14 and 16 are the server node claims corresponding to method claims 1-4 and 6, and are rejected under the same reasons set forth and the combination further teaches processing circuitry (e.g. Sheller, in paragraph 26, “code executed and/or data stored at the aggregator device”; Tomsett, in paragraphs 75-76, “programs 511, are stored on one or more of the computer readable storage media 508 for execution by one or more of the processors”).
As per claim 15, the rejection of claim 11 is incorporated and the combination further teaches wherein the ML model is a federated learning model (e.g. Sheller, in paragraph 13, “Federated and/or distributed learning enables a model to be trained using data across many edge systems”).
Claims 19 is the medium claim corresponding to method claim 1, and is rejected under the same reasons set forth and the combination further teaches a non-transitory computer readable storage medium storing a computer program (e.g. Sheller, in paragraph 26, “code executed and/or data stored at the aggregator device”; Tomsett, in paragraphs 75-76, “programs 511, are stored on one or more of the computer readable storage media 508 for execution by one or more of the processors”).
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (US 20190042878 A1) in view of Tomsett et al. (US 20200402658 A1) and Alistarh et al. ("Byzantine Stochastic Gradient Descent", arXiv:1803.08917v1, March 23 2018, pages 1-20) and further in view of Niwa et al. (US 20210158226 A1).
As per claim 5, the rejection of claim 1 is incorporated, but does not specifically teach wherein the first logical explanation for the unvalidated comprises a first equality or a first inequality for a first variable, and the second logical explanation comprises a second equality or a second inequality for the first. However, Niwa teaches a first logical explanation comprising a first equality or a first inequality for a first variable and a second logical explanation comprising a second equality or a second inequality for the first variable (e.g. in paragraphs 9, 17, 23, and 42, “the variables consist of three kinds of primal variables w.sub.1, w.sub.2, w.sub.3, six kinds of dual variables λ.sub.1|2, λ.sub.2|1, λ.sub.1|3, λ.sub.3|1, λ.sub.2|3, λ.sub.3|2, and six kinds of primal variables w.sub.1.sup.<1>, w.sub.1.sup.<2>, w.sub.2.sup.<1>, w.sub.2.sup.<2>, w.sub.3.sup.<1>, w.sub.3.sup.<2> exchanged between the nodes… x.sub.i=[x.sub.i,1, . . . , x.sub.i,B].sup.T, w.sub.i=[w.sub.i,1, . . . , w.sub.i,B].sup.T, and λ.sub.i|j=[λ.sub.i|j,1, . . . , λ.sub.i|j,B].sup.T ”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Niwa because one of ordinary skill in the art would have recognized the benefit of determining relevant information.
Claims 7 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (US 20190042878 A1) in view of Tomsett et al. (US 20200402658 A1) and Alistarh et al. ("Byzantine Stochastic Gradient Descent", arXiv:1803.08917v1, March 23 2018, pages 1-20) and further in view of Ignatiev et al. (“Abduction-Based Explanations for Machine Learning Models”, 11/26/2018, 10 pages).
As per claim 7, the rejection of claim 6 is incorporated and the combination further teaches wherein the first locally trained ML model is a neural network and the obtaining of the logical explanations includes information of the first locally trained ML model (e.g. Sheller. in paragraphs 47 and 63, “provides a model update to the example aggregator 110. In some examples, additional information is provided along with the model update such as, for example, an identity of the edge device 130, an indication of how much training data was used to prepare the model update, and/or other parameters identified as part of the model training process... provides model information concerning the private layers of the model (e.g., model updates resulting from training by the private model trainer 320)... may additionally provide model information concerning the public layers of the model… model updates originating from non-trusted edge devices”), but does not specifically teach the information including a logical encoding into mixed integer linear programming, wherein the logical explanations are a minimal set of input features that guarantee respective outputs. However, Ignatiev teaches logical explanations including logical encoding of neural networks into mixed integer linear programming, wherein logical explanations are a minimal set of input features that guarantee respective outputs (e.g. in pages 1-2, “our method provides formal guarantees on the generated explanations. For example, we can generate cardinality-minimal explanations… encoding of NNs into Mixed Integer Linear Programming is used”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Ignatiev because one of ordinary skill in the art would have recognized the benefit of improving model results.
Claim 17 is the server node claim corresponding to method claim 7, and is rejected under the same reasons set forth.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (US 20190042878 A1) in view of Tomsett et al. (US 20200402658 A1) and Alistarh et al. ("Byzantine Stochastic Gradient Descent", arXiv:1803.08917v1, March 23 2018, pages 1-20) and further in view of Morris et al. (US 20160350671 A1) and Fowler et al. (US 6493436 B1).
As per claim 8, the rejection of claim 6 is incorporated and the combination further teaches wherein the ML model predicts (e.g. Tomsett, in paragraph 30, “analytical model building based on sample data known as training data in order to make predictions or decisions”),
but does not specifically teach whether an equipment of a radio station is going to fail during a next predetermined interval, wherein the data stored in the client devices are maintenance records of equipment with operational parameter histories including failures.
However, Morris teaches a model predicting whether an equipment is going to fail during a next predetermined interval and data stored being maintenance records of equipment with operational parameter histories including failures (e.g. in paragraphs 32 and 75, “one might have relevant data on [equipment] (e.g., failure history) with a desire to predict the probability of [equipment] failure within a 1-hour prediction window… source data can include...maintenance records… a failure of a component or piece of equipment”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Morris because one of ordinary skill in the art would have recognized the benefit of using models for well-known applications,
but does not specifically teach equipment of a radio station.
However, Fowler teaches equipment of a radio station (e.g. in column 1 lines 49-55, “local radio station providing the audio having an equipment failure”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Fowler because one of ordinary skill in the art would have recognized the benefit of incorporating other well-known types of equipment (also amounts a simple substitution that yields predictable results [e.g. see KSR Int'l Co v. Teleflex Inc., 550 US 398,82 USPQ2d 1385,1396 (U.S. 2007) and MPEP 2143(B)]).
Claim 18 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (US 20190042878 A1) in view of Tomsett et al. (US 20200402658 A1), Alistarh et al. ("Byzantine Stochastic Gradient Descent", arXiv:1803.08917v1, March 23 2018, pages 1-20) and Ignatiev et al. (“Abduction-Based Explanations for Machine Learning Models”, 11/26/2018, 10 pages) and further in view of Morris et al. (US 20160350671 A1) and Fowler et al. (US 6493436 B1).
As per claim 18, the rejection of claim 17 is incorporated and the combination further teaches wherein the ML model predicts (e.g. Tomsett, in paragraph 30, “analytical model building based on sample data known as training data in order to make predictions or decisions”),
but does not specifically teach whether an equipment of a radio station is going to fail during a next predetermined interval, wherein the data stored in the client devices are maintenance records of equipment with operational parameter histories including failures.
However, Morris teaches a model predicting whether an equipment is going to fail during a next predetermined interval and data stored being maintenance records of equipment with operational parameter histories including failures (e.g. in paragraphs 32 and 75, “one might have relevant data on [equipment] (e.g., failure history) with a desire to predict the probability of [equipment] failure within a 1-hour prediction window… source data can include...maintenance records… a failure of a component or piece of equipment”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Morris because one of ordinary skill in the art would have recognized the benefit of using models for well-known applications,
but does not specifically teach equipment of a radio station.
However, Fowler teaches equipment of a radio station (e.g. in column 1 lines 49-55, “local radio station providing the audio having an equipment failure”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Fowler because one of ordinary skill in the art would have recognized the benefit of incorporating other well-known types of equipment (also amounts a simple substitution that yields predictable results [e.g. see KSR Int'l Co v. Teleflex Inc., 550 US 398,82 USPQ2d 1385,1396 (U.S. 2007) and MPEP 2143(B)]).
Claims 10, 20, and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (US 20190042878 A1) in view of Tomsett et al. (US 20200402658 A1) and Alistarh et al. ("Byzantine Stochastic Gradient Descent", arXiv:1803.08917v1, March 23 2018, pages 1-20) and further in view of Morris et al. (US 20160350671 A1).
Claim 10 corresponds to claim 6, and is rejected under the same reasons set forth, but the combination does not specifically teach wherein the local data includes maintenance records. However, Morris teaches maintenance records (e.g. in paragraphs 32 and 75, “one might have relevant data on [equipment] (e.g., failure history) with a desire to predict the probability of [equipment] failure within a 1-hour prediction window… source data can include...maintenance records… a failure of a component or piece of equipment”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of the combination to include the teachings of Morris because one of ordinary skill in the art would have recognized the benefit of using models for well-known applications.
Claims 20 is the medium claim corresponding to method claim 10, and is rejected under the same reasons set forth and the combination further teaches a non-transitory computer readable storage medium storing a computer program (e.g. Sheller, in paragraph 26, “code executed and/or data stored at the aggregator device”; Tomsett, in paragraphs 75-76, “programs 511, are stored on one or more of the computer readable storage media 508 for execution by one or more of the processors”).
Claim 23 the server node claim corresponding to method claim 10, and is rejected under the same reasons set forth and the combination further teaches processing circuitry (e.g. Sheller, in paragraph 26, “code executed and/or data stored at the aggregator device”; Tomsett, in paragraphs 75-76, “programs 511, are stored on one or more of the computer readable storage media 508 for execution by one or more of the processors”).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
For example,
De Brouwer et al. (US 20200293887 A1) teaches information including a distance based on updated information, the distance measuring a deviation of a respective ML model (e.g. in paragraph 163, “On the FL aggregator side, the federated learner can be configured to filter out spurious updates by calculating a distance measure that compares each modified tensor received from the edge devices to the base model tensor, constructing a distribution of distance measures in an updating cycle and rejecting from aggregation with the current version of outlier modified tensors. That is, production of the new base model version, will not be based on rejected tensors having a distance measure that are outliers from the distribution. An outlier can be determined using a statistical measure such as three standard deviations or the like”).
Choudhary et al. (US 20190385043 A1) teaches “local machine learning models 116a-116n correspond to the global machine learning model 108. For instance, in certain embodiments, the local machine learning models 116a-116n represent copies of the global machine learning model 108. In multiple training iterations, some or all of the client devices 112a-112n implement global parameters from the server(s) 102 in their respective local machine learning models 116a-116n and adjust the global parameters to reduce a loss determined locally by the client devices 112a-112n. Through multiple training iterations, the asynchronous training system 106 learns and incrementally adjusts the global parameters by receiving modified parameter indicators from a subset of the client devices 112a-112n and adjusting the global parameters at the server(s) 102 based on the modified parameter indicators received in each training iteration” (e.g. in paragraph 54).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WILLIAM WONG whose telephone number is (571)270-1399. The examiner can normally be reached Monday-Friday 9am-5pm.
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.
/W.W/Examiner, Art Unit 2144 06/27/2026
/SHOURJO DASGUPTA/Primary Examiner, Art Unit 2144
Prosecution Insights