CTNF 18/484,949 CTNF 94262 DETAILED ACTION This non-final office action is responsive to application 18/484,949 as submitted 11 Oct. 2023. Claim status is currently pending and under examination for claims 1-11 of which independent claims are 1 and 10-11. Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Priority Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55. The application has an effective filing date of 02/24/23. Information Disclosure Statement As required by MPEP 609(c), the applicant’s submissions of the Information Disclosure Statements dated 10/11/23 – 01/26/24 are acknowledged by the examiner and the cited references have been considered in the examination of the claims now pending. As required by MPEP 609 C(2), a copy of the PTOL-1449 initialed and dated by the examiner is attached to the instant office action. Specification The specification is objected to because the title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed, see MPEP 606.01. Claim Objections Claims 1, 4 and 10-11 are objected to because of the following informalities: claim limitations recite “in a case where” which is a contingent limitation (a case may be not where) and should read to positively include the limitation, see MPEP 2111.04. Appropriate correction or clarification is requested. Claim Rejections - 35 USC § 102 07-06 AIA 15-10-15 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. 07-07-aia AIA 07-07 The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – 07-08-aia AIA (a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention. 07-12-aia AIA (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. 07-15 AIA Claim s 1, 3-4, 6-8 and 10-11 are rejected under 35 U.S.C. 102( a)(1 ) as being anticipated by : Nguyen et al., “Where to Begin? Exploring the Impact of Pre-training and Initialization in Federated Learning” hereinafter Nguyen (arXiv: 2206.15387v1, Meta AI) . With respect to claim 1 , Nguyen teaches: A learning system comprising a plurality of local devices and a server {Nguyen [P.4] Alg.1 FedOPT (fed)erated learning (opt)imization with plurality of client devices & server, [P.1 ¶1-2]} , the plurality of local devices each comprising processing circuitry {Nguyen [P.13 SectA.2] implementation “machines, each with eight Nvidia V100 GPUs” GPU circuitry} configured to determine a federated local training condition indicating a training condition in federated learning of a local model based on preliminary local training information including a preliminary local training condition and a preliminary local training result in a case where a model is preliminarily trained using local data {Nguyen [P.4] Alg.1 L.5-6 “Initialize local model y 0 i <- x t -1 . Each client performs E epochs of local updates via y i k +1 = C LIENT O PT ( y i k , F i , η c ). Let y i E denote the result” initialization (preliminary) comprises pre-training per [P.5 Sect4.2 ¶3], [P.6 Sect5.1 ¶1] “pre-trained initialization.” The claimed conditions may regard hyperparameter tuning [P.6 Sect4.4] “hyperparameter sweeps to tune client and server learning rates” similar at [P.13 SectA.2] e.g. Alg.1 L.6 C LIENT O PT ( y i k , F i , η c ) where η c is client learning rate/step size. Datasets used and tested are described [P.4 Sect4.1], [P.13 SectA.1] Tbl.2]} , and the server comprising processing circuitry configured to determine a global training condition of a global model based on the preliminary local training information {Nguyen [P.4] Alg.1 L.10 “Server updates global model x t = S ERVER O PT ( x t -1 , -Δ t , η s , t )” based on aggregate L.9 sent from client L.7 upon local training L.6 that starts from preliminary initialized model Line5 as noted above, [P.5 Sect4.2 ¶3]. A global condition may include η s server learning rate/step size Alg.1 L.10, [P.6 Sect4.4]. The training may employ e.g. [P.5 Sect4.3] “SGD the server updates the global model using stochastic gradient descent” and circuitry requisite for implementation by GPU is disclosed [P.13 SectA.2]} . With respect to claim 3 , Nguyen teaches the system according to claim 1, wherein the preliminary local training condition includes information regarding at least one of a learning rate, a regularization, a mini-batch size, an initialization method, an optimizer, and selection of an architecture structure, regarding the local model {Nguyen Alg.1 L.6 described [P.4 Sect.3 Last¶] “training using C LIENT O PT with client learning rate” learning rate for local training with client optimizer, further tunable per [P.6 Sect4.4]. See also [P.5 Sect4.2] initialization strategies as well as [P.13 A.2] “batch size… select the best hyperparameters” i.e. architecture structure} . With respect to claim 4 , Nguyen teaches the system according to claim 1, wherein the preliminary local training result includes information regarding at least one of a recognition rate, a loss, a learning curve, a weight of a model, and a pruning result, in a case where the local model is trained using the preliminary local training condition {Nguyen Alg.1 L.6 [P.4 ¶1] Eq.1 “function F measures the average loss of a model with parameters w on the i th client’s training data, and p i > 0 is the weight given to client i ” teaches a loss as well as weights of a model, the loss function F is included in the C LIENT O PT function Alg.1 L.6 for a local training update with result y } . With respect to claim 6 , Nguyen teaches the system according to claim 1, wherein the server includes the processing circuitry further configured to: group, as a local device group, a plurality of local devices each including a similar data distribution or training situation based on the preliminary local training information {Nguyen [P.4¶2] “cohort of clients the each client in the cohort performs E epochs of training” cohort is group, the use of ‘or’ is read as either, a training ‘situation’ is Alg.1 where preliminary initialized model is updated locally by said clients for the subsequent global model update by server. See also [P.9 ¶1] “cosine similarity between client updates”} ; determine information regarding the local device group as the global training condition {Nguyen [P.4] Alg.1 L.7-10 determined information comprises update Δ sent from clients trained as a precondition for server to perform global model updates conditioned on the cohort of local clients} ; and perform the federated learning for each of the local device group based on the global training condition {Nguyen [P.4 ¶1] Eq.1 “federated training” hence (Fed)Opt Alg.1 where each client L.6 of the cohort of local clients are optimized with loss function F at Eq.1 for further global model update L.10} . With respect to claim 7 , Nguyen teaches the system according to claim 1, wherein the preliminary local training information includes a number of training epochs {Nguyen [P.4] Alg.1 L.1,6-7 “local epochs E …Each client performs E epochs of local updates …local training” described [P.4 ¶2]} , and the server includes the processing circuitry further configured to set a number of times of update of the global model based on the number of training epochs {Nguyen [P.4] Alg.1 L.1-2,10 “rounds T” outer for-loop “for each round t = 1, …, T do” cont’d L.10 server updates global model with S ERVER O PT ( x t -1 , -Δ t , η s , t ), [P.4 ¶2] “At round t , a server sends its model x to cohort of clients then each client cohort performs E epochs of training… t is the round number” e.g. [P.6 Sect4.4] “1000 rounds of training” Figs 1, 4} . With respect to claim 8 , Nguyen teaches the system according to claim 1, wherein each of the local device includes the processing circuitry further configured to correct the federated local training condition based on the preliminary local training information and a training progress of the federated learning {Nguyen [P.4] Eq.1 corrects by minimizing min f (w) over model loss function F such that Alg.1 L.6 C LIENT O PT includes F for local training updates which progresses over epochs and rounds of the FedOpt federated learning} . With respect to claim 10 , the rejection of claim 1 is incorporated. The difference in scope being a method of a learning system including limitations of system claim 1. Nguyen discloses [P.2 ¶1] “federated optimization methods” using Alg.1 [P.4]. The remainder of this claim is rejected for the same rationale as claim 1. With respect to claim 11 , the rejection of claim 1 is incorporated. The difference in scope being a non-transitory computer readable medium including instructions executable by processor to perform limitations of system claim 1. Nguyen discloses [P.13 A.2] “We implemented all algorithms in Pytorch” and [P.13 Ft.Nt] “Our code is available at https://github.com/facebookresearch/FLSim” software with hardware GPU processors known to include memory as requisite computer environment to carry out algorithm functionality. The remainder of this claim is rejected for the same rationale as claim 1 . Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. 07-20-aia AIA The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. 07-23-aia AIA The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: 1. Determining the scope and contents of the prior art. 2. Ascertaining the differences between the prior art and the claims at issue. 3. Resolving the level of ordinary skill in the pertinent art. 4. Considering objective evidence present in the application indicating obviousness or nonobviousness. 07-21-aia AIA Claim s 2 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Nguyen in view of Tian et al., “F ED BERT: When Federated Learning Meets Pre-Training” hereinafter Tian . With respect to claim 2 , Nguyen teaches the system according to claim 1, wherein the server includes the processing circuitry further configured to integrate a parameter of the local model received from each of the local devices and update the global model based on the global training condition {Nguyen [P.4] Alg.1 L.9-10 Server aggregate, aggregate is integrate, parameter denoted Δ weighted by p and aggregates ∑ i ∈ s over clients sending L.7 to server for global model update with respective conditions of S ERVER O PT function [P.4 Sect.4]} , and each of the local devices includes the processing circuitry further configured to Nguyen further teaches updates for local and global models Alg.1 L.6,10 however it is not readily apparent that the local update is ‘based on’ global update. Tian teaches: update the local model based on a parameter regarding the updated global model {Tian see [P.11-12] Alg.2-3 ClientUpdate() with local epochs and Server to Compute global model, notably [P.8 ¶1] “The server aggregates the updates and sends back the global BERT model to all clients in each round” similar [P.9 Last¶] “the server will aggregate N clients’ Transformer blocks to a global Transformer block. Then the global Transformer block is shared to all the clients” conveys that client local models are updated subject to global model update being sent back, e.g. Fig 4 sequential training and Fig 1-left client to server to client back again} . Tian is directed to federated learning with pre-training and initialization thus being analogous. A person having ordinary skill in the art would have considered it obvious prior to the effective filing date to update local model based on updated global model per Tian in combination to arrive at the invention as claimed for a motivation [P.3 Last2¶] “collaboratively train a pre-trained model for NLP with advanced federated learning… aggregate the information from the clients and update the Transformer layers during training and communication” which “can well support all devices or clients without high computation resources” and/or [P.22 Last2¶] “Our approach has the advantage of resource-constrained resource requirement and privacy-preserving as well as keeps the state-of-the-art performance.” With respect to claim 9 , Nguyen teaches the system according to claim 1. Tian teaches wherein the server includes the processing circuitry further configured to correct the global training condition based on the preliminary local training information and a training progress of the federated learning {Tian [P.11-12] Algs.2-3 Server with back propagation, Fig 6 illustrates. The backprop calculates gradients for updating global model collaborating with local clients that progresses over epochs and rounds of training for federated learning, corrects by backpropagation} . Motivation for combination is applied similarly as in claim 2 . 07-21-aia AIA Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Nguyen in view of Yang et al., US PG Pub No 2024/0112039A1 hereinafter Yang, as evidenced by Provisional 63/411,034 . With respect to claim 5 , Nguyen teaches the system according to claim 1. Yang teaches wherein the global training condition includes information of an individualization layer of a local model based on a difference in data distribution between the local devices {Yang see [0101] Eq.2 describes “normalization layers… local BN layers” are layers of a local model for training clients with a federated global model, see e.g. Fig 8:804 local BN statistics. In Eq.2 the subtraction is difference and distributions regard running means and variances [0102]. See also uniform standard deviation, uniform distribution and distance-based feature loss [0103-07]. Provisional support comprises [0099-100], Fig 8:804} . Yang is directed to federated learning thus being analogous. A person having ordinary skill in the art would have considered it obvious prior to the effective filing date to include layer of local model based on difference per Yang in combination to arrive at the invention as claimed as applying known techniques to known methods ready for improvement to yield predictable results and/or for a motivation of federated domain generalization with sampling from uniform distribution [0099,0105]. Provisional support comprises [0099-0106] . 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure : Hou et al., “Privately Customizing PreFineTuning to Better Match User Data in Federated Learning” arXiv: 2302.09042v1 Carnegie Mellon Fig 1, Alg.1 PreFineTuning Reddi et al., US Patent 12,271,810B2 Google - FedOpt Alg. 1,3 Sidahmed et al., US PG Pub No 2023/0214642A1 Google – Alg. [0047] ServerOpt/ClientOpt Du et al., “Rethinking Normalization in Federated Learning” layerNorm, groupNorm >> BN Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to Chase P Hinckley whose telephone number is (571)272-7935. The examiner can normally be reached M-F 9:00 - 5:00. 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, Miranda M. Huang can be reached at 571-270-7092. 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. /CHASE P. HINCKLEY/Examiner, Art Unit 2124 Application/Control Number: 18/484,949 Page 2 Art Unit: 2124 Application/Control Number: 18/484,949 Page 3 Art Unit: 2124 Application/Control Number: 18/484,949 Page 4 Art Unit: 2124 Application/Control Number: 18/484,949 Page 5 Art Unit: 2124 Application/Control Number: 18/484,949 Page 6 Art Unit: 2124 Application/Control Number: 18/484,949 Page 7 Art Unit: 2124 Application/Control Number: 18/484,949 Page 8 Art Unit: 2124 Application/Control Number: 18/484,949 Page 9 Art Unit: 2124 Application/Control Number: 18/484,949 Page 10 Art Unit: 2124 Application/Control Number: 18/484,949 Page 11 Art Unit: 2124
