Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of the Claims
Claims 1, 10, and 13 have been amended. Claims 1-20 are currently pending and have been fully considered by the Examiner.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are 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.
In claim 1, the final 2 lines render the claim indefinite because it is unclear what this limitation means as a whole, and it is unclear what is being characterized, and it is unclear if “algorithm selection” means selecting one of the candidate training algorithms. Specification paragraph [0023] discloses that “CASH” stands for Combined Algorithm Selection and HyperParameter (HP) Optimization. It is unclear how the algorithm selection as recited in claim 1 is related to CASH. Examiner treats the final 2 lines of claim 1 to mean selecting a candidate training algorithm prior to federated learning (FL).
Claims 2-9 are rejected for failing to cure the deficiencies of claim 1.
Claims 10 recites the same indefinite limitation as method claim 1 and is therefore rejected for at least the same reasons.
Claims 11-12 are rejected for failing to cure the deficiencies of claim 10.
Claim 13 is a computer system which recites the same indefinite limitation as method claim 1 and is therefore rejected for at least the same reasons.
Claims 14-20 are rejected for failing to cure the deficiencies of claim 13.
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.
Claims 1-4, 6-7, 9-18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Gao et al. (“FIFL: A Fair Incentive Mechanism for Federated Learning,” cited in PTO-892 issued 05/01/2026) in view of Cacheda Seijo et al. (US 20120310770 A1, cited in PTO-892 issued 05/01/2026), hereinafter “Cacheda”, and Zhou et al. (“Single-shot Hyper-parameter Optimization for Federated Learning: A General Algorithm & Analysis,” cited in Applicant’s IDS filed 7/30/2023).
Regarding claim 1, Gao teaches: A method comprising: for at least a first two iterations: for a plurality of candidate training algorithms, obtaining, at a federated learning aggregator, from each of a plurality of federated learning clients, a score corresponding to a best hyperparameter, based on
for the plurality of candidate algorithms, the federated learning aggregator aggregating the obtained scores and (All of § 3.1 starting at page 2, col. 2 to page 3, col. 1 discloses federated learning. Equation 2 discloses aggregating the local gradients (“the obtained scores”).)
for the plurality of subsequent iterations:
for a best-performing subset of the plurality of candidate training algorithms, obtaining, at the federated learning aggregator, from each of the plurality of federated learning clients, an updated best hyperparameter and an updated corresponding score, based on calculates a contribution threshold bh which effectively prevents workers with low utility from joining the federation. The feature “a best-performing subset of the plurality of candidate training algorithms” are the models at workers with positive contribution (high utility), and an “updated best parameter and an updated corresponding score… as compared respectively to… a previous one of the subsequent iterations” is a local gradient.)
for the best-performing subset of the plurality of candidate algorithms, the federated learning aggregator aggregating the obtained updated scores
characterized in that algorithm selection (CASH) is fully completed prior to federated learning (FL). (Examiner treats this limitation to mean selecting a candidate training algorithm prior to federated learning. Page 6, col. 1, § 4.5, lines 5-6 discloses all workers upload their models (“candidate training algorithms)” before the FL training.)
However, Gao does not explicitly teach: for at least a first two iterations: obtaining a score, based on a fraction of available training data;
updating a projected score for each of the candidate training algorithms; and
increasing the fraction of available training data to be used in a subsequent one of the at least a first two iterations or a first one of a plurality of subsequent iterations;
for the plurality of subsequent iterations: obtaining an updated best hyperparameter and an updated corresponding score, based on a further increased fraction of available training data
and further updating a projected score for each of the best-performing subset of the candidate training algorithms;
But Cacheda teaches: for at least a first two iterations: [training]
[training] based on a further increased fraction of available training data ([0039], lines 1-14, [0046] and Fig. 6 discloses the percentage of available data to be used in the training set ranges from 40% to 90% over multiple tests.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incrementally increased Gao’s available training data according to Cacheda’s schedule. A motivation for the combination is that a small percentage allows evaluating the method under sparsity conditions, common in the initial phases, or in domains with a large number of users and/or items. With high percentages of ratings in the training set, the behavior of the method under relatively high density conditions can be evaluated. (Cacheda, [0039])
However, Gao and Cacheda do not explicitly teach: updating a projected score for each of the candidate training algorithms; and
and further updating a projected score for each of the best-performing subset of the candidate training algorithms;
But Zhou teaches: updating a projected score for each of the candidate training algorithms; and (Page 3, § 3, lines 1-8 and Page 5, § 3.2, lines 1-7 discloses a federated learning system for a machine learning model which generates a unified loss surface. A projected score is a unified loss surface, and each candidate training algorithm is a local model at each party or client.)
and further updating a projected score for each of the
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have generated Zhou’s loss surface for the clients of Gao and Cacheda. In the combination of references, a loss surface would be generated for clients with high contributions (the “best performing subset”) during subsequent iterations. A motivation for the combination is to identify a single set of good hyper-parameters that are subsequently used in a single FL training, in order to enable federated learning hyper-parameter optimization with minimal additional communication overhead. (Zhou, Abstract)
Regarding claim 2, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 1,
Gao teaches: further comprising the aggregator communicating with the clients to cooperatively train a federated global model based on a best one of the best-performing subset of the candidate training algorithms and its corresponding hyperparameters. (On page 3, col. 1, Equation 3 discloses learning parameters of global model F via federated learning. On page 5, Equations 13 and 14 calculate a worker’s contribution. A best one of the best-performing subset is a model of a worker with the highest contribution C.)
Regarding claim 3, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 2,
However, Gao and Zhou do not explicitly teach: wherein the plurality of subsequent iterations are continued until the fraction of available training data reaches 100%.
But Cacheda teaches: wherein the plurality of subsequent iterations are continued until the fraction of available training data reaches 100%. ([0039], lines 10-14 discloses the evaluation subset is composed by randomly selecting 10% of the dataset, while the training subset increases to 90% of the dataset. This corresponds to 100% of available training data as claimed.)
A motivation for the combination is the same as the motivation given for claim 1.
Regarding claim 4, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 2,
Gao teaches: further comprising carrying out inferencing with the trained federated global model. (On page 8, Fig. 8(b) teaches a test loss of the global FL model trained on CIFAR10 data set. A test loss indicates carrying out inference.)
Cacheda at [0039], lines 1-3 and 12-13 teaches evaluating a trained model using an evaluation subset of data.
Regarding claim 6, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 1,
Gao teaches: wherein the at least first two iterations comprise a first three iterations. (Page 3 , col. 1, lines 7-9 discloses local iteration “t”. Since Gao does not place an upper limit, the number of local iterations can reach at least 3. On page 8, Fig. 8 shows at least 1,800 communication iterations.)
Regarding claim 7, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 1, wherein the federated learning aggregator aggregating the obtained updated scores and further updating a projected score for each of the best-performing subset of the candidate training algorithms comprises
Gao teaches: one of max voting and averaging a best hyperparameter setting of each of the clients. (On page 3, col. 1, Equation 2 calculates a global gradient by performing a weighted sum of the local gradients for each worker i. A weighted sum is an average. Page 5, col. 2, last sentence of § 4.3 discloses workers with low utilities are prevented from joining the federation. A best hyperparameter setting is calculated from workers with high utilities.)
Regarding claim 9, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 1, wherein the federated learning aggregator aggregating the obtained updated scores and further updating a projected score for each of the best-performing subset of the candidate training algorithms comprises
However, Gao and Cacheda do not explicitly teach: performing regression using hyperparameters from each of the clients and their losses to generate top-K hyperparameter settings for re-evaluation.
But Zhou teaches: performing regression using hyperparameters from each of the clients and their losses to generate top-K hyperparameter settings for re-evaluation. (Page 4, § 3.1, lines 7-11 discloses a federated learning aggregator collecting all attempted pairs of hyperparameters and associated losses for a machine learning algorithm A from each of a plurality of parties (clients). On page 5, § 3.2, lines 1-8 and page 6, subsection “MPLM” teaches generating top-1 hyperparameter settings per client. With respect to the feature of re-evaluation, Page 10, subsection “Implementation” discloses evaluating the final performance using validation data.)
It would have been obvious to have incorporated Zhou’s regression into the combination of Gao, Cacheda, and Zhou. A motivation for the combination is the same as the motivation given for claim 1.
Regarding claim 10, Gao teaches: for at least a first two iterations: for a plurality of candidate training algorithms, obtaining, at a federated learning aggregator, from each of a plurality of federated learning clients, a score corresponding to a best hyperparameter, based on
for the plurality of candidate algorithms, the federated learning aggregator aggregating the obtained scores and (All of § 3.1 starting at page 2, col. 2 to page 3, col. 1 discloses federated learning. Equation 2 discloses aggregating the local gradients (“the obtained scores”).)
for a best-performing subset of the plurality of candidate training algorithms, obtaining, at the federated learning aggregator, from each of the plurality of federated learning clients, an updated best hyperparameter and an updated corresponding score, based on h which effectively prevents workers with low utility from joining the federation. The feature “a best-performing subset of the plurality of candidate training algorithms” are the models at workers with positive contribution (high utility), and an “updated best parameter and an updated corresponding score… as compared respectively to… a previous one of the subsequent iterations” is a local gradient.)
for the best-performing subset of the plurality of candidate algorithms, the federated learning aggregator aggregating the obtained updated scores
characterized in that algorithm selection (CASH) is fully completed prior to federated learning (FL). (Examiner treats this limitation to mean selecting a candidate training algorithm prior to federated learning. Page 6, col. 1, § 4.5, lines 5-6 discloses all workers upload their models (“candidate training algorithms)” before the FL training.)
However, Gao does not explicitly teach: A computer program product for implementing a federated learning aggregator on a computer, the computer program product comprising: a computer readable storage medium having stored thereon: first program instructions executable by the computer to cause the computer to, for at least a first two iterations: obtaining a score, based on a fraction of available training data;
updating a projected score for each of the candidate training algorithms; and
increasing the fraction of available training data to be used in a subsequent one of the at least a first two iterations or a first one of a plurality of subsequent iterations; and
second program instructions executable by the computer to cause the computer to, for the plurality of subsequent iterations:
obtaining an updated best hyperparameter and an updated corresponding score, based on a further increased fraction of available training data
and further updating a projected score for each of the best-performing subset of the candidate training algorithms;
But Cacheda teaches: A computer program product for implementing [software]
[training]
second program instructions executable by the computer to cause the computer to, ([0024], lines 10-14)
[training] based on a further increased fraction of available training data ([0039], lines 1-14, [0046] and Fig. 6 discloses the percentage of available data to be used in the training set ranges from 40% to 90% over multiple tests.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incrementally increased Gao’s available training data according to Cacheda’s schedule, and to have incorporated Cacheda’s computer containing a processor and memory comprising instructions into Gao. A motivation for the combination is that a small percentage allows evaluating the method under sparsity conditions, common in the initial phases, or in domains with a large number of users and/or items. With high percentages of ratings in the training set, the behavior of the method under relatively high density conditions can be evaluated (Cacheda, [0039]). A motivation for incorporating Cacheda’s computer is to execute a federated learning aggregator in the real world.
However, Gao and Cacheda do not explicitly teach: updating a projected score for each of the candidate training algorithms; and
and further updating a projected score for each of the best-performing subset of the candidate training algorithms;
But Zhou teaches: updating a projected score for each of the candidate training algorithms; and (Page 3, § 3, lines 1-8 and Page 5, § 3.2, lines 1-7 discloses a federated learning system for a machine learning model which generates a unified loss surface. A projected score is a unified loss surface, and each candidate training algorithm is a local model at each party or client.)
and further updating a projected score for each of the
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have generated Zhou’s loss surface for the clients of Gao and Cacheda. In the combination of references, a loss surface would be generated for clients with high contributions (the “best performing subset”) during subsequent iterations. A motivation for the combination is to identify a single set of good hyper-parameters that are subsequently used in a single FL training, in order to enable federated learning hyper-parameter optimization with minimal additional communication overhead. (Zhou, Abstract)
Regarding claim 11, the combination of Gao, Cacheda, and Zhou teaches: The computer program product of Claim 10,
Gao teaches: further comprising
However, Gao and Zhou do not explicitly teach: third program instructions executable by the computer to cause the computer
But Cacheda teaches: third program instructions executable by the computer to cause the computer ([0024], lines 10-14)
A motivation for the combination is the same as the motivation given for claim 10.
Claim 12 recites a computer program product which implements the same features as the method of claim 3 and is therefore rejected for at least the same reasons.
Claim 13 recites a computer system which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
Gao teaches a federated learning aggregator as a server at page 3, col. 1, in the sentence above Equation 2, and col. 2, lines 1-5. However, Gao does not explicitly teach that the federated learning aggregator comprises a memory; and at least one processor, coupled to the memory, and operative to [execute operations].
But Cacheda teaches: a memory, and at least one processor, coupled to the memory, and operative to execute operations. ([0024], lines 10-14 and 20-23)
It would have been obvious to a person having ordinary skill in the art to have incorporated Cacheda’s computer containing a processor and memory into Gao. A motivation for the combination is to execute a federated learning aggregator in the real world.
Claim 14-18 and 20 each recites a computer system which implements the same features as the method of claims 2-4, 6-7, and 9, respectively, and are therefore rejected for at least the same reasons.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Gao et al. (“FIFL: A Fair Incentive Mechanism for Federated Learning,” cited in PTO-892 issued 05/01/2026) in view of Cacheda Seijo et al. (US 20120310770 A1, cited in PTO-892 issued 05/01/2026), hereinafter “Cacheda”, Zhou et al. (“Single-shot Hyper-parameter Optimization for Federated Learning: A General Algorithm & Analysis,” cited in Applicant’s IDS filed 7/30/2023), and Tuor et al. (US 20210158099 A1, cited in PTO-892 issued 05/01/2026).
Regarding claim 5, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 4,
However, Gao, Cacheda, and Zhou do not explicitly teach: wherein the clients comprise thin clients.
But Tuor teaches: wherein the clients comprise thin clients. ([0029], lines 1-6 and [0031], from line 1 to “a thin client” in line 7 discloses a federated learning server interconnected with thin clients via a network.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Tuor’s thin clients into the combination of Gao, Cacheda, and Zhou. A motivation for the combination is that an end user of a thin client does not manage or control the underlying cloud infrastructure including network, servers, operating systems, storage, or even individual application capabilities, with the possible exception of limited user-specific application configuration settings. (Tuor, [0087])
Claims 8 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Gao et al. (“FIFL: A Fair Incentive Mechanism for Federated Learning,” cited in PTO-892 issued 05/01/2026) in view of Cacheda Seijo et al. (US 20120310770 A1, cited in PTO-892 issued 05/01/2026), hereinafter “Cacheda”, Zhou et al. (“Single-shot Hyper-parameter Optimization for Federated Learning: A General Algorithm & Analysis,” cited in Applicant’s IDS filed 7/30/2023), and Lin et al. (US 20240104367 A1, cited in PTO-892 issued 05/01/2026).
Regarding claim 8, the combination of Gao, Cacheda, and Zhou teaches: The method of Claim 1, wherein the federated learning aggregator aggregating the obtained updated scores and further updating a projected score for each of the best-performing subset of the candidate training algorithms comprises
However, Gao, Cacheda, and Zhou do not explicitly teach: taking a union of a top K hyperparameter settings of each of said clients, which is sent for re-evaluation.
But Lin teaches: taking a union of a top K hyperparameter settings of each of said clients, which is sent for re-evaluation. (All of [0025] and [0040] discloses the global machine learning model is a union of different orthogonal model partitions. Each partition comprises parameters (hyperparameter settings). Taking a union of a top K hyperparameter settings amounts to combining the updated model partitions from the selected client devices, which were selected for having the best computing performances. Sending for re-evaluation amounts to deploying the updated machine learning model to the client devices for use in performing inferences at each of the client devices.)
It would have been obvious to a person having ordinary skill in the art to have incorporated Lin’s model partitions into the combination of Gao, Cacheda, and Zhou. A motivation for the combination is that each partition may be independently updated without adversely affecting the performance of other partitions in the machine learning model. (Lin, [0025])
Claim 19 recites a computer system which implements the same features as the method of claim 8 and is therefore rejected for at least the same reasons.
Response to Arguments
On page 9 in the remarks filed 08/03/2026, the Applicant argues, “Gao does not fairly disclose or suggest that algorithm selection is fully completed prior to federated learning. In short, Gao is simply not concerned with reconciling the CASH problem with the HPO problem in federated learning contexts. Actually, Gao repeatedly indicates the opposite – that is, the more conventional approach whereby federated learning occurs throughout the process (refer, e.g., to at least sections 3.1 and 4.3 discussing FL integration).”
Applicant's arguments have been fully considered but they are not persuasive. Examiner treats the limitation in the final 2 lines of claim 1 to mean selecting a candidate training algorithm prior to federated learning. In Gao, page 6, col. 1, § 4.5, lines 5-6 discloses all workers upload their models before the FL training, where Gao’s models correspond to “candidate training algorithms” as claimed.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Asher H. Jablon whose telephone number is (571)270-7648. The examiner can normally be reached Monday - Friday, 9:00 am - 6:00 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar can be reached at (571)270-3169. 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.
/A.H.J./Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Prosecution Insights